WO2017045640A1

WO2017045640A1 - Associated stream bandwidth scheduling method and apparatus in data center

Info

Publication number: WO2017045640A1
Application number: PCT/CN2016/099209
Authority: WO
Inventors: 郭宏翔; 于佳; 张东旭; 安高峰; 廖屏
Original assignee: 中兴通讯股份有限公司
Priority date: 2015-09-18
Filing date: 2016-09-18
Publication date: 2017-03-23
Also published as: CN106549782A

Abstract

Disclosed in the embodiment are an associated stream bandwidth scheduling method and apparatus in a data center, comprising: a master controller selecting the bottleneck data stream of each associated stream according to a collected associated stream bandwidth request; the master controller allocating for each bottleneck data stream a transmission bandwidth that meets a first condition, and determining, on the basis of the transmission bandwidths allocated to the bottleneck data streams, the estimated completion time of the associated streams to which the bottleneck data streams belong; and the master controller allocating transmission bandwidths for non-bottleneck data streams in the associated streams according to the estimated completion time of the associated streams, and when the transmission bandwidths allocated for the non-bottleneck data streams in all the associated streams meet a second condition, determining the estimated completion time of the associated streams and the transmission bandwidths allocated for the data streams in each associated stream. The associated stream bandwidth scheduling method and apparatus in a data center, disclosed in the embodiment, can combine with the associated streams used in the data center to realize dynamic bandwidth scheduling for an all-optical network of an optical burst switching (OBS) data center.

Description

Bandwidth scheduling method and device for associated stream in data center

Technical field

The present invention relates to the field of network communication technologies, and in particular, to a bandwidth scheduling method and apparatus for an associated flow in a data center.

Background technique

Information services represented by mobile Internet and cloud computing are increasingly relying on high-performance, scalable data centers. Usually the data center carries many user-oriented applications such as web servers, file services, online games, enterprise applications, etc., as well as computationally intensive tasks such as big data mining. These application tasks result in a large number of concurrent communication flows between servers (between virtual machines). These communication data streams include one-to-many, many-to-one, many-to-many communication modes, and have a wide dynamic range of bandwidth requirements. Special delay requirements are required to ensure the performance of the upper application. Therefore, how to design the corresponding data exchange bearer network, support diverse applications and complex communication modes in the data center, and provide the advantages of greener and lower energy consumption and the ability to flexibly upgrade and expand are the hot topics of common concern in industry and academia. . Especially in recent years, with the development of optoelectronic device technology, data center network design based on optical transmission and exchange has received more and more attention and attention.

Within the data center, most cluster computing application frameworks, such as MapReduce, perform user-defined work and are transported along specific workflows that conform to the programming model. Others are user-facing channels where users request and ultimately return corresponding answers (such as Google and Bing's search results, as well as Facebook's home page feedback).

Take MapReduce's Shuffle and Distributed File System (DFS) replication process as an example. MapReduce is a well-known and widely used distributed computing framework. In this model, each mapper reads input from DFS, performs user-defined calculations, and writes intermediate data to disk; each reducer reads intermediate data from different mappers, merges them, and writes their output to DFS, then copy to multiple destinations. The main feature of the MapReduce model is that until the last reducer is completed, the entire task is carry out. Therefore, at the end of the task there is a clear barrier that researchers have used in the model to optimize the Shuffle process. Another example is the Bux Synchronous Parallel (BSP) model, which is another well-known model in cluster computing. The computational framework using this model has Pregel, Giragh, and Hama focused on graph processing, matrix computing, and network algorithms. A BSP parallel computer consists of a set of processors - memory units interconnected by a communication network. It has three main parts: a set of distributed processors with local memory, a global data communication network, and a mechanism to support global barrier synchronization between all processing units. The superstep communication phase can be optimized by optimizing the last obstacle for each superstep. In another example, in the aggregation process in Partition-aggregate communication, the user-oriented online service receives the user's request and passes it through the aggregation tree to the following worker nodes. At each stage of the tree, independent requests in different segmentation processes. The activity is generated. Finally, the worker responds to the aggregation and returns to the user interface at the deadline. The response that cannot be returned at the deadline is discarded or sent asynchronously after a while (such as Facebook home feedback).

As can be seen from the above description, although the streams are indistinguishable from each other in the transport layer, in the same cluster computing, streams between different groups of computers typically have application-level semantic relevance. For example, the last stream in the MapReduce Shuffle process determines the completion time of the overall stream. Similarly, if a stream is delayed or discarded, it will cause the overall stream to miss the latest completion time, which may affect a small portion of the response. Therefore, a stream with semantic relevance between different sets of computers is referred to as an associated stream. In other words, an associated flow is a collection of flows with the same performance goal. The performance goal can be to ensure that the flow can be completed before the deadline, or to have the shortest transmission time.

One of the key technologies for implementing data center all-optical networks based on optical burst switching is to provide a dynamic resource scheduling mechanism with reliable and application characteristics. The flexible bandwidth allocation algorithm can realize flexible scheduling of network bandwidth resources to meet the requirements. The dynamic bandwidth requirement of the network node to which the request is applied. However, for the data center optical burst switching network, the prior art network bandwidth resource scheduling mechanism does not consider the associated stream demand feature of the upper layer application.

Summary of the invention

The embodiments of the present invention provide a bandwidth scheduling method and device for an associated flow in a data center, which can implement dynamic bandwidth scheduling of an optical optical all-optical network in an optical burst exchange according to at least an associated flow applied in the data center.

In order to achieve at least the above technical purpose, an embodiment of the present invention provides a bandwidth scheduling method for an associated flow in a data center, including: the primary controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request; The controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determines an expected completion time of the associated flow of each bottleneck data flow according to the transmission bandwidth allocated to each bottleneck data stream; the main controller completes according to the expected completion of each associated flow. The time allocates a transmission bandwidth for the non-bottleneck data stream in each associated stream. When the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition, the estimated completion time of each associated flow is determined and allocated to each associated flow. The transmission bandwidth of each data stream.

Optionally, before the primary controller selects the bottleneck data flow of each associated flow according to the collected associated flow bandwidth request, the method further includes: the primary controller collecting all associated flow bandwidth requests in one cycle.

Optionally, the associated flow bandwidth request includes source address information, destination address information, and amount of data to be transmitted of each data flow in the associated flow.

Optionally, the associated flow bandwidth request further includes an upper limit value of the completion time of the associated flow, where an estimated completion time of the associated flow corresponding to the determined bottleneck data flow does not exceed an upper limit of the completion time of the associated flow value.

Optionally, the primary controller selects, according to the collected associated flow bandwidth request, the bottleneck data flow of each associated flow, where the primary controller requests, according to each associated flow bandwidth request, each data flow information. Among all the data streams of the associated stream, the data stream with the largest amount of data to be transmitted or the data stream with the largest available bandwidth of the node is selected as the bottleneck data stream of the associated stream.

Optionally, after the primary controller allocates transmission bandwidth to the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, the method further includes: when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows is not satisfied In the second condition, the main controller uniformly reduces the transmission bandwidth allocated to each bottleneck data stream that satisfies the first condition, according to the re-allocated bottleneck data. The transmission bandwidth of the flow re-determines the estimated completion time of the associated flow of each bottleneck data flow, and then allocates the transmission bandwidth for the non-bottleneck data flow in each associated flow according to the expected completion time of each associated flow until the non-bottleneck is allocated to all associated flows. The transmission bandwidth of the data stream satisfies the second condition.

Optionally, the first condition includes: a transmission bandwidth of each bottleneck data stream does not exceed a maximum line rate of the sender end and the receiver end server; and a sum of transmission bandwidths of all bottleneck data streams having the same source node does not exceed the source node. The maximum available bandwidth, the sum of the transmission bandwidths of all bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node.

Optionally, the second condition includes: a transmission bandwidth of each non-bottleneck data stream does not exceed a maximum line rate of the sending end and the receiving end server; and a sum of transmission bandwidths of all non-bottleneck data streams having the same source node does not exceed The maximum available bandwidth of the source node, the sum of the transmission bandwidths of all non-bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node; the transmission bandwidth allocated to the non-bottleneck data stream in an associated flow is not less than the associated flow The expected completion time determines the transmission bandwidth of the non-bottleneck data stream.

Optionally, after the primary controller determines an expected completion time of each associated flow and a transmission bandwidth allocated to each data flow in each associated flow, the method further includes: the primary controller transmitting a transmission cycle of each associated flow and each The transmission bandwidth of each data stream in the associated flows is sent to each network node.

In addition, the embodiment of the present invention further provides a bandwidth scheduling apparatus for an associated flow in a data center, which is disposed on the main controller, and includes: a flow classification module, configured to select each associated flow according to the collected associated flow bandwidth request. The bottleneck data stream; the expected completion time calculation module is configured to allocate a transmission bandwidth satisfying the first condition for each bottleneck data stream, and determine an expected completion time of the associated stream to which each bottleneck data stream belongs according to the transmission bandwidth allocated to each bottleneck data stream. The associated allocation module is configured to allocate a transmission bandwidth for the non-bottleneck data stream in each associated stream according to the expected completion time of each associated stream, and determine that each of the associated non-bottleneck data streams in the associated stream meets the second condition. The estimated completion time of the associated flow and the transmission bandwidth allocated to each data flow in each associated flow.

According to still another embodiment of the present invention, a storage medium is also provided. The storage medium is arranged to store program code for performing the following steps:

The main controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request;

The primary controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determines an expected completion time of the associated flow to which each bottleneck data stream belongs according to a transmission bandwidth allocated to each bottleneck data stream;

The primary controller allocates a transmission bandwidth for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and determines the associated flow when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition. The estimated completion time and the transmission bandwidth allocated to each data stream in each associated stream.

In the embodiment of the present invention, the primary controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request; the primary controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data flow, and according to the allocation The estimated bandwidth of each bottleneck data stream belongs to the transmission bandwidth of each bottleneck data stream; the main controller allocates the transmission bandwidth for the non-bottleneck data stream in each associated stream according to the estimated completion time of each associated stream, when all the associations are allocated When the transmission bandwidth of the non-bottleneck data stream in the flow satisfies the second condition, the estimated completion time of each associated flow and the transmission bandwidth allocated to each data flow in each associated flow are determined. As such, the embodiment of the present invention implements bandwidth dynamic scheduling of the optical burst all-optical network in the optical burst exchange in combination with the associated flow applied in the data center.

Moreover, through the embodiments of the present invention, an efficient and collision-free dynamic resource scheduling of an optical burst transmission ring network in a data center is realized, and an application-based fast data transmission is realized, which not only can allocate bandwidth resources fairly and reasonably, and quickly respond to bursts. The bandwidth requirement of the service, and the data collision-free exchange, the overall transmission completion time of the associated flow task is short, and a high bandwidth utilization is obtained.

DRAWINGS

The drawings described herein are intended to provide a further understanding of the invention, and are intended to be a part of the invention. In the drawing:

FIG. 1 is a schematic diagram of an application scenario according to an embodiment of the present invention;

2 is a flowchart of a bandwidth scheduling method for an associated flow in a data center according to an embodiment of the present invention;

3 is a specific flowchart of step 11 in the embodiment of the present invention;

4 is a specific flowchart of step 12 in the embodiment of the present invention;

FIG. 5 is a specific flowchart of step 13 in the embodiment of the present invention;

FIG. 6 is a schematic diagram of a bandwidth scheduling apparatus for an associated flow in a data center according to an embodiment of the present invention.

detailed description

The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

It is to be understood that the terms "first", "second" and the like in the specification and claims of the present invention are used to distinguish similar objects, and are not necessarily used to describe a particular order or order.

FIG. 1 is a schematic diagram of an application scenario according to an embodiment of the present invention. As shown in FIG. 1, the optical burst transmission ring network in the data center includes n network nodes and one main controller. Where n is an integer greater than 1, and the number of network nodes can be set as desired. The main controller is responsible for centralized control management such as bandwidth allocation. Each network node is a rack configured with m servers in the data center (in fact, the number of servers can be set as needed), and each network node is configured with k transmissions. And k receivers, where m and k are integers greater than one.

Here, the optical burst transmission network in the data center adopts a time connection between network nodes, and all connection establishment times are equal in one cycle, and network nodes can be equivalent to full connection, and the associated flow is sent by the server. The line rate is a Gbps, and each network node includes k wavelengths. Each wavelength transmission rate is b Gbps. The maximum output bandwidth of each network node is k*b Gbps, and the maximum receiving bandwidth is k*b Gbps. The maximum transmission bandwidth supported between any two network nodes in the period is k*b/(n-1) Gbps. The associated stream data is transmitted on the link in the form of a stream.

FIG. 2 is a flowchart of a bandwidth scheduling method for an associated flow in a data center according to an embodiment of the present invention; Figure. As shown in FIG. 2, the bandwidth scheduling method for the associated flow in the data center provided by this embodiment includes the following steps:

Step 11: The primary controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request.

Here, before step 11, the method further includes: the main controller collecting all associated stream bandwidth requests in one cycle.

The associated stream bandwidth request includes source address information, destination address information, and amount of data to be transmitted of each data stream in the associated stream. Optionally, the associated flow bandwidth request further includes an upper limit of the completion time of the associated flow.

Specifically, during each bandwidth allocation period, the server proposes an associated flow bandwidth request to the primary controller through an associated flow application programming interface (API). Since the bandwidth scheduling is performed in units of one cycle, the primary control is performed. The device counts all associated stream bandwidth requests in a cycle.

In this case, the step 11 includes: the main controller selects, according to each data stream information carried in each associated stream bandwidth request, the data stream with the largest amount of data to be transmitted or the data with the largest available bandwidth of the node from all the data streams of the associated stream. The stream acts as a bottleneck data stream for the associated stream.

For example, as shown in FIG. 3, for each associated stream bandwidth request, all the data streams in the associated stream carried by the associated stream are sorted according to the amount of data to be transmitted, and the data with the largest amount of data to be transmitted is selected. The stream with the largest available bandwidth of the stream or node serves as the element of the bottleneck data stream matrix, and the non-bottleneck data stream acts as the element of the non-bottleneck data stream matrix, where the non-bottleneck data stream is the data stream other than the bottleneck data stream in the associated stream. If the associated stream does not select the bottleneck data stream, continue to select the bottleneck data stream for the associated stream; when all the associated streams select the bottleneck data stream, the final bottleneck data stream matrix and the non-bottleneck data stream are output. matrix. In this case, the bottleneck data flow matrix is composed of the data flows with the largest amount of data to be transmitted among all the associated flows collected, and the completion time of the associated flows can be initially determined by calculating the transmission time of the bottleneck data flow of each associated flow. The non-bottleneck data flow matrix is used as the final decision for the completion time of the associated stream and the bandwidth allocation basis for all data streams.

Step 12: The primary controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determines an expected completion time of the associated flow to which each bottleneck data stream belongs according to the transmission bandwidth allocated to each bottleneck data stream.

The first condition includes: the transmission bandwidth of each bottleneck data stream does not exceed the maximum line rate of the sender end and the receiver end server (eg, 10 Gbps); and the sum of the transmission bandwidths of all bottleneck data streams having the same source node does not exceed the source node. The maximum available bandwidth, the sum of the transmission bandwidths of all bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node. Optionally, the estimated completion time of the associated flow of the bottleneck data flow does not exceed the upper limit of the completion time of the associated flow carried by the associated flow bandwidth request.

Specifically, as shown in FIG. 4, for the bottleneck data flow matrix, all bottleneck data streams are classified according to the same source node and having the same destination node, and each bottleneck data stream is allocated according to 10 Gbps bandwidth, and then added. The amount of bandwidth occupied by the associated stream data that has not been transmitted in the previous cycle. The bottleneck data stream with the same source node in the next cycle occupies the total bandwidth of the source node, and the bottleneck data stream with the same destination node occupies the total bandwidth of the destination node. According to the maximum bandwidth allocation method, it is determined whether the current bandwidth is sufficient. After that, it is determined whether the calculated total amount of the occupied source node bandwidth and the total occupied node bandwidth exceed the maximum value (such as k*b Gbps), and if any of the values exceeds the maximum value, the data is uniformly reduced and allocated to each bottleneck data. The transmission bandwidth of the stream (for example, reduced on a 10 Gbps basis) until the calculated total amount of occupied source node bandwidth and the total occupied node bandwidth do not exceed the maximum value. If the total bandwidth of the occupied source node and the total bandwidth of the occupied destination node calculated in this step do not exceed the maximum value, the transmission bandwidth allocated for each bottleneck data stream and the data volume to be transmitted of each bottleneck data stream are respectively calculated. The transmission completion time of the bottleneck data stream of each associated flow, and the transmission completion time is the estimated completion time of the associated flow.

Step 13: The primary controller allocates a transmission bandwidth for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and determines the association when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition. The estimated completion time of the flow and the transmission bandwidth allocated to each data flow in each associated flow.

In this case, after the primary controller allocates transmission bandwidth to the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, the method further includes: when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows is not satisfied. In the second condition, the main controller uniformly reduces the transmission bandwidth allocated to each bottleneck data stream that satisfies the first condition, and re-determines the estimated completion time of the associated flow of each bottleneck data stream according to the re-allocated transmission bandwidth of each bottleneck data stream, and then The transmission bandwidth is re-allocated for the non-bottleneck data stream in each associated flow according to the expected completion time of each associated flow until the transmission bandwidth allocated to the non-bottleneck data flow in all associated flows satisfies the second condition.

The second condition includes: the transmission bandwidth of each non-bottleneck data stream does not exceed the maximum line rate of the sender end and the receiver end server (eg, 10 Gbps); the sum of the transmission bandwidths of all non-bottleneck data streams having the same source node does not exceed The maximum available bandwidth of the source node, the sum of the transmission bandwidths of all non-bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node; the transmission bandwidth allocated to the non-bottleneck data stream in an associated flow is not less than the associated flow The expected completion time determines the transmission bandwidth of the non-bottleneck data stream.

Specifically, as shown in FIG. 5, for a non-bottleneck data stream matrix, all non-bottleneck data streams are classified according to the same source node and having the same destination node, wherein the transmission bandwidth allocated for each non-bottleneck data stream is respectively performed. The estimated completion time of the associated flow calculated in step 12 and the corresponding amount of data to be transmitted may be determined, and the amount of bandwidth occupied by the associated flow data that has not been transmitted in the previous cycle is added, and the next cycle has the same source node. The non-bottleneck data stream occupies the total bandwidth of the source node, and the non-bottleneck data stream with the same destination node occupies the total bandwidth of the destination node. After that, it is determined whether the calculated total bandwidth of the occupied source node and the total occupied node bandwidth exceed the maximum value (such as k*b Gbps), and if any of the values exceeds the maximum value, the data is uniformly reduced and allocated to each bottleneck data. The transmission bandwidth of the stream is reduced according to the transmission bandwidth of the bottleneck data stream determined in step 12, and returns to step 12 to re-determine the estimated completion time of each associated stream according to the transmission bandwidth of the reduced bottleneck data stream, and then follow the steps. 13 The judgment is made until the total amount of occupied source node bandwidth and the total occupied node bandwidth in step 13 do not exceed the maximum value. After the adjustments of

steps

12 and 13 are performed, when all the associated flows can be successfully transmitted according to the determined expected completion time, the transmission bandwidth finally allocated for each data flow in each associated flow and the prediction of each associated flow are obtained. Complete time.

Optionally, the estimated completion time of the associated flow of the bottleneck data flow does not exceed the upper limit of the completion time of the associated flow carried by the associated flow bandwidth request.

If the transmission of one or some data flows cannot be completed within the upper limit of the completion time due to the conflict, the associated flow of the data flow is postponed as a whole, so that the start transmission time of the associated flow is delayed. One or several bandwidth scheduling periods until a point in time is reached that is calculated according to the above steps so that all data streams can be completed within the upper limit of the completion time.

Here, after step 13, the method further includes: the main controller transmitting the transmission period of each associated stream and the transmission bandwidth of each data stream in each associated stream to each network node.

In summary, the selection of the bottleneck data flow in the associated flow is implemented through step 11; the estimated completion time of the associated flow is initially determined through step 12; and the estimated completion time of the associated flow is revised through step 13, so that the associated flow collected in one cycle is obtained. Applications can be successfully transmitted at the expected completion time.

In addition, the embodiment of the present invention further provides a bandwidth scheduling apparatus for an associated flow in a data center, which is disposed on the main controller, and includes: a flow classification module, configured to select each associated flow according to the collected associated flow bandwidth request. a bottleneck data stream; an estimated completion time calculation module configured to allocate a transmission bandwidth satisfying the first condition for each bottleneck data stream, and determine an expected completion time of the associated stream to which each bottleneck data stream belongs according to a transmission bandwidth allocated to each bottleneck data stream; The association allocation module is configured to allocate a transmission bandwidth for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and determine the association when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition. The estimated completion time of the flow and the transmission bandwidth allocated to each data flow in each associated flow.

Further, the apparatus further includes: an input module configured to collect all associated stream bandwidth requests within one cycle.

Further, the flow classification module is specifically configured to: according to each data flow information carried in each associated flow bandwidth request, select the data flow with the largest amount of data to be transmitted or the largest available bandwidth of the node from all the data flows of the associated flow. The data stream acts as a bottleneck data stream for the associated stream.

Further, the association allocation module is also set to be the number of non-bottlenecks assigned to all associated flows When the transmission bandwidth of the flow does not satisfy the second condition, the transmission bandwidth that is allocated to each bottleneck data stream that satisfies the first condition is uniformly reduced, and the reduced transmission bandwidth of each bottleneck data stream is fed back to the estimated completion time. And a calculation module, wherein the estimated completion time calculation module re-determines an estimated completion time of the associated flow to which each bottleneck data stream belongs.

The first condition includes: the transmission bandwidth of each bottleneck data stream does not exceed the maximum line rate of the sender end and the receiver end server; the sum of the transmission bandwidths of all bottleneck data streams having the same source node does not exceed the maximum available bandwidth of the source node. The sum of the transmission bandwidths of all bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node.

The second condition includes: the transmission bandwidth of each non-bottleneck data stream does not exceed the maximum line rate of the sender end and the sink end server; the sum of the transmission bandwidths of all non-bottleneck data streams having the same source node does not exceed the maximum of the source node. Available bandwidth, the sum of the transmission bandwidths of all non-bottleneck data streams with the same destination node does not exceed the maximum available bandwidth of the destination node; the transmission bandwidth allocated to the non-bottleneck data stream in an associated flow is not less than the estimated completion time according to the associated flow Determine the transmission bandwidth of the non-bottleneck data stream.

Further, the apparatus further includes: an output module, configured to send a transmission period of each associated stream and a transmission bandwidth of each data stream in each associated stream to each network node.

FIG. 6 is a schematic diagram of a bandwidth scheduling apparatus for an associated flow in a data center according to an embodiment of the present invention. As shown in FIG. 6, the bandwidth scheduling apparatus for the associated flow in the data center provided by the embodiment includes: a system status monitoring module 301, a bandwidth adjustment module 302, an input module 303, a flow classification module 304, an estimated completion time calculation module 305, and an association. The module 306 is distributed and the module 307 is output.

The system status monitoring module 301 is configured to monitor the system status, for example, including the occupancy of any two nodes when transmitting the associated stream, and send the monitoring result to the bandwidth adjustment module 302;

The bandwidth adjustment module 302 is configured to adjust the bandwidth between any two nodes of the network. Specifically, the bandwidth adjustment module 302 includes an optical layer bandwidth calculation module and a network state database, and the optical layer bandwidth calculation module is monitored according to the system status monitoring module 301. Network status, the optical layer bandwidth utilization can be calculated, and the result is sent to the network status database for the estimated completion time calculation. Module 305 queries the current optical layer bandwidth configuration;

The input module 303 is configured to implement the generation of the associated stream bandwidth request matrix. Specifically, the input module 303 collects the bandwidth request from the server and organizes it into an associated stream bandwidth request matrix, where the content includes the data stream number, the source node of the request, and the destination. Node and amount of data to be transmitted;

The flow classification module 304 is configured to implement a calculation process from the associated flow bandwidth request matrix to the bottleneck data flow matrix. Specifically, the flow classification module 304 sorts each associated flow bandwidth request matrix according to the amount of data to be transmitted, and selects each Generating a bottleneck data stream matrix by associating a data stream with the largest amount of data to be transmitted in the associated stream;

The expected completion time calculation module 305 is configured to implement a calculation process from the bottleneck data flow matrix to the expected completion time. Specifically, the expected completion time calculation module 305 determines any associated flow request based on the existing bandwidth resource and the actual associated flow bandwidth request. The completion time calculation module 305 includes, for example, a first bandwidth amount calculation unit, a first flow reduction unit, and an estimated completion time calculation unit, wherein the first bandwidth amount calculation unit is configured to allocate data to each bottleneck. The bandwidth of the stream and the amount of bandwidth occupied by the associated stream data that has not been transmitted in the previous period. The bottleneck data stream with the same source node in the next cycle occupies the total bandwidth of the source node, and the bottleneck data stream with the same destination node occupies the total bandwidth of the destination node. The first stream reduction unit is configured to determine whether the result calculated by the first bandwidth amount calculation unit exceeds a corresponding maximum value, and if exceeded, uniformly reduce the transmission bandwidth allocated to each bottleneck data stream, and if not exceeded, the expected completion time The calculation unit is based on the transmission assigned to each bottleneck data stream Width, and the amount of data to be transmitted is calculated for each data flow bottleneck in the transmission time, i.e., the bottleneck associated with each data stream belongs projected completion time of the stream;

The association allocation module 306 is configured to implement a mapping process from the estimated completion time of the associated flow to the non-bottleneck data flow bandwidth allocation of the associated flow. Specifically, the estimated completion time of the associated flow obtained by the estimated completion time calculation module 305 is associated. The non-bottleneck data stream in the flow allocates a transmission bandwidth. The association allocation module 306 includes, for example, a second bandwidth amount calculation unit, a second stream reduction unit, and a feedback unit, where the second bandwidth amount calculation unit is configured to be configured according to each associated flow. The estimated completion time allocates bandwidth to the non-bottleneck data stream, and calculates the next source with the same source according to the bandwidth allocated to the non-bottleneck data stream and the amount of bandwidth occupied by the associated stream data that has not been transmitted in the previous period. The non-bottleneck data stream of the node occupies the total bandwidth of the source node, and the non-bottleneck data stream with the same destination node occupies the total bandwidth of the destination node, and the second stream reduction unit is configured to determine whether the result calculated by the second bandwidth amount calculation unit exceeds the corresponding The maximum value, if exceeded, uniformly reduces the transmission bandwidth allocated to each bottleneck data stream, and the feedback unit feeds the reduced result back to the expected completion time calculation module 35, and the estimated completion time calculation module 35 re-determines the associated flow. The expected completion time, if not exceeded, the estimated completion time of each associated stream and the transmission bandwidth allocated to each data stream is sent to the output module 307;

The output module 307 is configured to implement the generation of the associated stream bandwidth map, specifically, to collect the bandwidth allocation information obtained from the association allocation module 306 and the estimated completion time calculation module 305, and organize the data into an associated stream bandwidth map, the content including the association. Streaming cycle, amount of bandwidth occupied.

In an actual application, the association allocation module 306 and the estimated completion time calculation module 305 are two iterative modules. When the input load is too large, the bandwidth of the data stream in each associated stream needs to be continuously adjusted to collect the data. The associated streams can match the bandwidth transmission.

It should be noted that the device shown in FIG. 6 can be used alone when the system state monitoring module 301 and the bandwidth adjustment module 302 are not needed.

Embodiments of the present invention also provide a storage medium. Optionally, in the embodiment, the foregoing storage medium may be configured to store program code for performing the following steps:

S1. The primary controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request.

S2, the primary controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determines an expected completion time of the associated flow to which each bottleneck data stream belongs according to a transmission bandwidth allocated to each bottleneck data stream;

S3. The main controller allocates a transmission bandwidth for the non-bottleneck data stream in each associated stream according to an estimated completion time of each associated stream. When the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition, determining each The estimated completion time of the associated flow and the transmission bandwidth allocated to each data flow in each associated flow.

Optionally, in this embodiment, the foregoing storage medium may include, but not limited to, a USB flash drive, a Read-Only Memory (ROM), a Random Access Memory (RAM), a mobile hard disk, and a magnetic memory. A variety of media that can store program code, such as a disc or a disc.

For example, the specific examples in this embodiment may refer to the examples described in the foregoing embodiments and the optional embodiments, and details are not described herein again.

In practical applications, the above modules may be implemented by a processor executing programs/instructions stored in a memory. However, the present invention is not limited thereto, and the functions of the above modules/units may also be through firmware/logic circuits/integrated circuits. achieve.

It will be apparent to those skilled in the art that the various modules or steps of the present invention described above can be implemented by a general-purpose computing device that can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they may be implemented by program code executable by the computing device such that they may be stored in the storage device by the computing device and, in some cases, may be different from the order herein. The steps shown or described are performed, or they are separately fabricated into individual integrated circuit modules, or a plurality of modules or steps thereof are fabricated as a single integrated circuit module. Thus, the invention is not limited to any specific combination of hardware and software.

The above description is only the preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes can be made to the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Industrial applicability

As described above, the bandwidth scheduling method and apparatus for the associated flow in the data center provided by the embodiment of the present invention have the following beneficial effects: the embodiment of the present invention implements the optical burst exchange data center by combining the associated flow applied in the data center. Dynamic bandwidth scheduling of all-optical networks. Moreover, through the embodiments of the present invention, an efficient and collision-free dynamic resource scheduling of an optical burst transmission ring network in a data center is realized, and fast application-based data transmission is realized, which not only can allocate bandwidth resources reasonably and reasonably. And quickly respond to the bandwidth requirements of the burst service, and achieve data collision-free exchange, the associated transmission task overall transmission completion time is short and obtain a higher bandwidth utilization.

Claims

A bandwidth scheduling method for an associated flow in a data center, comprising:

The main controller selects a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request;

The primary controller allocates a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determines an expected completion time of the associated flow to which each bottleneck data stream belongs according to a transmission bandwidth allocated to each bottleneck data stream;

The primary controller allocates a transmission bandwidth for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and determines the associated flow when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition. The estimated completion time and the transmission bandwidth allocated to each data stream in each associated stream.
The method of claim 1, wherein the master controller selects a bottleneck data stream of each associated stream according to the collected associated stream bandwidth request, and further includes: the main controller collecting one cycle All associated stream bandwidth requests.
The method of claim 1 or 2, wherein the associated stream bandwidth request comprises source address information, destination address information, and amount of data to be transmitted for each data stream in the associated stream.
The method of claim 3, wherein the associated stream bandwidth request further comprises a completion time upper limit value of the associated flow, wherein an estimated completion time of the associated flow of the determined bottleneck data flow does not exceed the The upper limit of the completion time of the associated stream.
The method of claim 1, wherein the primary controller selects a bottleneck data stream of each associated flow according to the collected associated flow bandwidth request, the primary controller carries the request according to each associated flow bandwidth request. Each data stream information selects, from all data streams of the associated stream, a data stream with the largest amount of data to be transmitted or a data stream with the largest available bandwidth of the node as a bottleneck data stream of the associated stream.
The method of claim 1, wherein the primary controller allocates transmission bandwidth to the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and further includes: when the non-bottleneck is allocated to all associated flows The transmission bandwidth of the data stream does not satisfy the above In the second condition, the main controller uniformly reduces the transmission bandwidth that is allocated to each bottleneck data stream that satisfies the first condition, and re-determines the estimated completion time of the associated flow of each bottleneck data stream according to the re-allocated transmission bandwidth of each bottleneck data stream. Then, the transmission bandwidth is allocated for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, until the transmission bandwidth allocated to the non-bottleneck data flow in all associated flows satisfies the second condition.
The method according to claim 1 or 6, wherein said first condition comprises: a transmission bandwidth of each bottleneck data stream does not exceed a maximum line rate of the sender end and the sink end server; all bottleneck data streams having the same source node The sum of the transmission bandwidths does not exceed the maximum available bandwidth of the source node, and the sum of the transmission bandwidths of all bottleneck data streams having the same destination node does not exceed the maximum available bandwidth of the destination node.
The method according to claim 1 or 6, wherein said second condition comprises: a transmission bandwidth of each non-bottleneck data stream does not exceed a maximum line rate of the sender end and the sink end server; all non-bottlenecks having the same source node The sum of the transmission bandwidths of the data streams does not exceed the maximum available bandwidth of the source node, and the sum of the transmission bandwidths of all non-bottleneck data streams having the same destination node does not exceed the maximum available bandwidth of the destination node; the non-bottleneck data stream allocated to an associated stream The transmission bandwidth is not less than the transmission bandwidth of the non-bottleneck data stream determined according to the expected completion time of the associated flow.
The method of claim 1, wherein after the main controller determines an estimated completion time of each associated flow and a transmission bandwidth allocated to each data stream in each associated flow, the method further includes: the primary controller The transmission period of the associated stream and the transmission bandwidth of each data stream in each associated stream are sent to each network node.
A bandwidth scheduling device for an associated flow in a data center is disposed on a main controller, and includes:

a flow classification module, configured to select a bottleneck data flow of each associated flow according to the collected associated flow bandwidth request;

The estimated completion time calculation module is configured to allocate a transmission bandwidth that satisfies the first condition for each bottleneck data stream, and determine the number of bottlenecks according to the transmission bandwidth allocated to each bottleneck data stream. The estimated completion time of the associated flow according to the flow;

The association allocation module is configured to allocate a transmission bandwidth for the non-bottleneck data stream in each associated flow according to the estimated completion time of each associated flow, and determine the association when the transmission bandwidth allocated to the non-bottleneck data stream in all associated flows satisfies the second condition. The estimated completion time of the flow and the transmission bandwidth allocated to each data flow in each associated flow.
A storage medium, configured to store a computer program for performing a bandwidth scheduling method for an associative flow within a data center according to any one of claims 1-9.