CN104702694A - Dynamic data stream scheduling method of data center based on hybrid transmission mode - Google Patents

Abstract

Disclosed is a dynamic data stream scheduling method of a data center based on a hybrid transmission mode. The dynamic data stream scheduling method of the data center based on the hybrid transmission mode comprises, in a data center network based on the hybrid transmission mode, taking the volume of a data stream, a network topological graph formed by connecting switches in a wired mode, an interference graph formed by connecting the switches in a wireless mode and available wireless connection at input, solving a data stream scheduling problem by integrating the methods of relaxation algorithm, linear programming and the like, and lastly, obtaining a data stream transmission route and link service conditions as algorithm output. By reasonable scheduling mass data streams in the data center, the dynamic data stream scheduling method of data center based on the hybrid transmission mode eliminates the hotspot problem of the data center, solves jamming and improves the integral performance of the data center can be improved; meanwhile, the dynamic data stream scheduling method of data center based on the hybrid transmission mode can control the consumption within a certain range and accordingly avoid introducing new burdens to the data center. The dynamic data stream scheduling method of data center based on the hybrid transmission mode can reasonably schedule mass data streams in a network, eliminate hotspots and jamming of the data center network and improve the performance of the data center.

Description

Based on the dynamic data stream scheduling method of the data center of mixed transmission modes

Technical field

The invention belongs to Internet technical field, particularly a kind of dynamic data stream scheduling method of the data center based on mixed transmission modes.

Background technology

In recent years, rapidly, cloud computing technology obtains and advances greatly and improve in Internet technology development.As the infrastructure of cloud computing, the development of data center is to the propelling of cloud computing and universally bring unprecedented benefit.At present, the application service based on cloud computing presents rapid growth state, applies to the every field in living, brings great convenience to the work of people, study, amusement.But, along with the development of data center, and the growth of extensive service based on cloud computing technology, these are served the data flow produced and present polarization feature, the existing low-volume traffic stream to latency sensitive, have again exporting responsive high amount of traffic, to data center network, huge challenge is brought with scheduling to the route of these data flow.How efficiently to process the bottleneck that blended data stream becomes the development of restricting data central site network.

Combine to the low-volume traffic stream of delay-sensitive with to exporting responsive high amount of traffic, dynamically produce popular connection in heart network in the data, cause network congestion, data center's performance is had a strong impact on.In the recent period, have researcher to propose, data flow hot topic connected, by reasonably scheduling, is assigned in the lower connection of other utilization rate, can effective alleviating network congestion problem.

At present, the solution for network congestion problem is mainly divided into two types, the first, redesigns and realizes data center network topology structure, utilizes new topological property to solve network congestion and hot issue.The second, in the middle of traditional data center, add wireless portion, utilize the feature of wireless transmission two-forty, low delay, solve network congestion problem.

60GHz wireless communication technique is as a kind of emerging technology, have that transmission rate is high, strong interference immunity, fail safe be high, high directivity, permission transmitting power high, applied to traditional cable data center, formed the data center of mixed transmission modes.By reasonably disposing and dispatching, utilizing 60GHz technology to transmit high amount of traffic, a new thinking can be provided for solving network congestion problem.

Summary of the invention

In order to overcome the shortcoming of above-mentioned prior art, the object of the present invention is to provide a kind of dynamic data stream scheduling method of the data center based on mixed transmission modes, when ensureing that application program normally works, the mode of mixed logic dynamic is adopted in the data center of mixed transmission modes, high amount of traffic is dispatched, by utilizing wired connection and wireless connections, solving data center network congestion problems, promoting data center's overall performance.

To achieve these goals, the technical solution used in the present invention is:

Based on a dynamic data stream scheduling method for the data center of mixed transmission modes, realize according to following steps under the data center network of mixed transmission modes:

Step (1), by data flow f _ksize d ^k, the network topological diagram G that formed by wired connection of interchanger _w(V, E _w), the interference figure G that formed by wireless connections of interchanger _i(V _i, E _i) and available wireless connection as input, wherein, V represents interchanger set, E _wrepresent wired connection set, V _irepresent the set that the interchanger node that there is interference is mutually formed, E _irepresent the set that the wireless connections of interference are mutually formed, interference figure G _i(V _i, E _i) represent in interference relationships;

Step (2), utilizes linear programming to obtain the optimal Decomposition set of optimum link assignment problem

Step (3), for arbitrary interfering nodes v _i, v _i∈ V _i, according to with the optimal Decomposition set that step (2) obtains obtain the weights omega (v of its correspondence _i), wherein e _ijrepresent the limit connecting i and j, connect if existed, then e _iibe 1, if there is no, then e _ijbeing that 0, F represents handling object, is several data flow f _kthe set formed, λ represents link utilization, C _ijrepresent link capacity, ω (v _e) represent the weight of interfering nodes;

Step (4a), the multiple independent set IS after acquisition decomposition and the mark set { y (I of its correspondence _i);

Step (4b), according to mark Resource selection independent set IS that (4a) obtains, is referred to as I ^*;

Step (5), for arbitrarily for γ _ijassignment, if e _ij∈ I ^*, γ _ijbe 1, represent e _ijbe added into and separate set, otherwise γ _ij=0, represent e _ijbe not added into and separate set;

Step (6), in conjunction with γ obtained in the previous step _ij, upgrade { γ _ijset, obtain new { γ _ij, namely final { γ _ij, for { γ _ijin be 1 element, its represent connection belong to final disaggregation; In conjunction with final { γ _ijand the middle optimal Decomposition set obtained of step (2) obtain represent that whether a kth data flow is through i node and j node;

Step (7a), according to by data flow f _kbe assigned to different paths, obtain a set of paths, the requirement of each paths all demand fulfillment data flow;

Step (7b), for arbitrary data flow f _k, according to probability selection one paths, join set of paths;

Step (8), return path set { P ^kand { γ _ij, terminate computing.

Available wireless is connected in any limit, put it into interference figure G _i(V _i, E _i) as a node, if any two nodes exist interference, then between these two nodes, add a limit (e ₁, e ₂); For interference figure G _i(V _i, E _i) in node, if there is not limit between two nodes, then show that these two nodes are separate, independent set IS is interference figure G _i(V _i, E _i) a subset, in IS, any two nodes are all independent, in order to set up wireless connections, from interference figure G _i(V _i, E _i) a middle selection independent set IS, assuming that

${\mathrm{\γ}}_{\mathrm{ij}}=0,{\∀e}_{\mathrm{ij}}\∈\stackrel{\‾}{E_s}---\left(1\right)$

${\mathrm{\γ}}_{\mathrm{ij}}+{\mathrm{\γ}}_{\mathrm{uv}}\≤1,\∀(e_{\mathrm{ij}},e_{\mathrm{uv}})\∈E_I---\left(2\right)$

Wherein, for e _ij∈ E _w, γ _ijrepresent e _ijwhether selected, if choose, be then 1, otherwise be 0, if the connection existence interference that two selected, then cannot meet formula (2), that is, all selected connections are from same independent set IS.

Optimal Decomposition set in described step (2) method for solving be:

For C _ij

Solve

minλ

Belong to NP-hard problem, use relaxed algorithm by decomposition step, finally utilize linear programming to obtain the optimal Decomposition set of optimum link assignment problem that is, utilize Linear programming thought to obtain a mark set, the some points in the middle of corresponding convex combination, namely gather

for binary number, represent data flow f _kwhether through and connection e _ij,

${\mathrm{\Σ}}_{e_{\mathrm{ij}}\∈E_w}{x}_{\mathrm{ij}}^k\≤{\mathrm{\δ}}_1,\∀f_k\∈F---\left(3\right)$

${\mathrm{\Σ}}_{e_{\mathrm{ij}}\∈\stackrel{\‾}{E_s}}{x}_{\mathrm{ij}}^k\≤{\mathrm{\δ}}_2,\∀f_k\∈F---\left(4\right)$

Wherein, δ ₁, δ ₂represent the quantity of the wired and wireless connections in routed path.

Situation is taken in conjunction with the size of each data flow and every bar limit, and link capacity and link utilization, calculate e _ijweights omega (e _ij).

In described step (4a), utilize graph coloring algorithm, by interference figure G _iwith weight set ω (v _i) obtain interference figure G as algorithm input _idecomposition result { y (I _i): I _i∈ I}, the independent set IS after obtaining decomposing and the mark set { y (I of its correspondence _i).

In described step (4b), according to mark set y (I ^*) corresponding mark character, select the independent set IS that is optimum.

In described step (6), in conjunction with final { γ _ijand the middle optimal Decomposition set obtained of step (2) obtain if be 1, represent that a kth data flow through i node and j node, otherwise, without i node and j node.

In described step (7a), for if be 1, f _kthrough e _ij, otherwise, without it, final f _kobtain mulitpath, Stochastic choice one transmits.

Compared with prior art, the present invention, by approximate data and linear programming, solves Path Selection.This algorithm is applied to the data center of mixed transmission modes, effectively can solve hot issue, alleviate the pressure of wire transmission, improving data transmission efficiency.

Accompanying drawing explanation

Fig. 1 is schematic flow sheet of the present invention.

Embodiment

Embodiments of the present invention are described in detail below in conjunction with drawings and Examples.

A kind of dynamic dataflow scheduling scheme of the present invention, flexible in conjunction with the stable of wire transmission and wireless transmission, dispatch data stream, this invention is mainly used in the transmission solving high amount of traffic, alleviates the hot issue in data center network.In order to realize the transmission bandwidth of Gb rank, 60GHz wireless technology is used in the middle of the present invention, but this technology can produce a fan-shaped interference.

Handling object of the present invention is the set that several high amount of traffics are formed, assuming that this set is F, for arbitrary f _k, all there is d in ∈ F ^krepresent the size of this data flow.This invention G (V, E) represents blended data central site network, and V represents interchanger set, and E represents the set of paths connecting switch, E _wrepresent wired connection set, E _srepresent wireless connections set, can E=E be obtained _w+ E _s.In the present invention, represent available wireless connections set, interference figure G _i(V _i, E _i) represent in interference relationships.For in any limit, put it into G _i, as a node, if any two nodes exist interference, then between these two nodes, add a limit (e ₁, e ₂).

For the node in interference figure, if there is not limit between two nodes, then show that these two nodes are separate.Independent sets IS closes and represents interference figure G _ia subset, in the middle of IS, any two nodes are all independent.

In order to set up wireless connections, need to select an independent set from interference set, assuming that

${\mathrm{\γ}}_{\mathrm{ij}}=0,{\∀e}_{\mathrm{ij}}\∈\stackrel{\‾}{E_s}---\left(1\right)$

${\mathrm{\γ}}_{\mathrm{ij}}+{\mathrm{\γ}}_{\mathrm{uv}}\≤1,\∀(e_{\mathrm{ij}},e_{\mathrm{uv}})\∈E_I---\left(2\right)$

Wherein, γ _ijrepresent e _ij(e _ij∈ E _w) whether selected, if choose, be then 1, otherwise be 0.If two selected connection exists interference, then (2) are run counter to.So all selected connections need from same independent set IS.

In this invention, the wired connection in the middle of mixed logic dynamic and wireless connections can be controlled, use a binary number represent data flow f _kwhether through and connection e _ij, can obtain

${\mathrm{\Σ}}_{e_{\mathrm{ij}}\∈E_w}{x}_{\mathrm{ij}}^k\≤{\mathrm{\δ}}_1,\∀f_k\∈F---\left(3\right)$

${\mathrm{\Σ}}_{e_{\mathrm{ij}}\∈\stackrel{\‾}{E_s}}{x}_{\mathrm{ij}}^k\≤{\mathrm{\δ}}_2,\∀f_k\∈F---\left(4\right)$

Wherein, δ ₁, δ ₂represent the quantity of the wired and wireless connections in routed path.

For e _ij∈ E, represent link utilization, wherein C _ijrepresent link capacity.Need to solve a maximum link utilization λ, make namely minimum λ is required solution.

In order to the problem solved, obtain minimum λ, need to solve a linear programming problem P.By carrying out relaxation processes to former problem, utilizing Linear programming thought to obtain a mark set, the some points in the middle of corresponding convex combination, obtaining set

According to obtain solution, according to

$\mathrm{\ω}\left(v_e\right)=\mathrm{\ω}\left(e_{\mathrm{ij}}\right)=\frac{{\mathrm{\Σ}}_{f_k\∈F}{d}^k{x}_{\mathrm{ij}}^k}{\mathrm{\λ}{C}_{\mathrm{ij}}}---\left(5\right)$

Calculate any v _i∈ V _icorresponding ω (v _i).Afterwards, according to the decomposition result { y (I that Similarity algorithm obtains _i): I _i∈ I}, obtains the IS after decomposing.According to y (I ^*), select an independent set IS.

For arbitrarily if e _ij∈ I ^*, then γ _ijbe set to 1, otherwise be 0.According to new { γ _ij, obtain

Next need the problem that solution one is new, be about to decompose a set of paths { P ^k: f _k∈ F}, each path p should meet requirement D (p) of data flow.

To any f _k∈ F, according to probability obtain P ^*, then by P ^*add P ^k.

Finally return { P ^kand { γ _ij, algorithm terminates.

Step (1) starts this algorithm, by data flow f _ksize d ^k, wired network topology figure G _w(V, E _w), interference figure G _i(V _i, E _i), available wireless connects as algorithm input, wherein, V represents interchanger set, E _wrepresent wired connection set, G _w(V, E _w) represent the network topological diagram that interchanger is consisted of wired connection.For the interchanger node of wireless connections, due to the interference characteristic of 60GHza itself, can there is interference relationships between node, the node that there is interference mutually forms set V _i, the wireless connections of interference mutually form set E _i, interference figure G _ithen by V _iand E _icommon formation.

Step (2) solve maximum link utilization be this invention final solve target, for

$\underset{f_k\∈F}{\mathrm{\Σ}}{d}^k{x}_{\mathrm{ij}}^k\≤\mathrm{\λ}{C}_{\mathrm{ij}}$

Solve

minλ

Belong to NP-hard problem, we use relaxed algorithm to former problem, by decomposition step, utilize linear programming to obtain the optimal Decomposition set of optimum link assignment problem

Step (3) is for arbitrary interfering nodes v _i, according to

$\mathrm{\ω}\left(v_e\right)=\mathrm{\ω}\left(e_{\mathrm{ij}}\right)=\frac{{\mathrm{\Σ}}_{f_k\∈F}{d}^k{x}_{\mathrm{ij}}^k}{\mathrm{\λ}{C}_{\mathrm{ij}}}$

With the optimal Decomposition set that step (1) obtains obtain the weights omega (v of its correspondence _i), wherein e _ijrepresent the limit connecting i and j.

Weights omega (the v that the interfering nodes that the interference figure that algorithm inputs by step (4) and previous step obtain is corresponding _i) the set ω that forms as the input of colouring algorithm, obtain the mark set { y (I of independent set IS and its correspondence _i).According to y (I ^*) select an independent set I ^*.

Step (5) is for arbitrarily for γ _ijassignment.If e _ij∈ I ^*, γ _ijbe 1, otherwise γ _ij=0.γ _ijrepresent e _ijwhether be added into set, wherein represent the set that available wireless connections are formed.

The γ that step (6) obtains in conjunction with previous step _ij, upgrade { γ _ijset, obtain new { γ _ij, namely final { γ _ij.For { γ _ijbe 1 element, its represent connection belong to final disaggregation.In conjunction with { γ _ijand the middle optimal Decomposition set obtained of step (2) can obtain represent that whether a kth data flow is through i node and j node.

Step (7) basis by data flow f _kbe assigned to different paths, obtain a set of paths, the requirement of each paths all demand fulfillment data flow.For arbitrary data flow f _k, according to probability selection one paths, join set of paths.

Step (8) return path set { P ^kand { γ _ij, terminate computing.

Claims (9)

1. based on a dynamic data stream scheduling method for the data center of mixed transmission modes, it is characterized in that, realize according to following steps under the data center network of mixed transmission modes:

Step (1), by data flow f _ksize d ^k, the network topological diagram G that formed by wired connection of interchanger _w(V, E _w), the interference figure G that formed by wireless connections of interchanger _i(V _i, E _i) and available wireless connection as input, wherein, V represents interchanger set, E _wrepresent wired connection set, V _irepresent the set that the interchanger node that there is interference is mutually formed, E _irepresent the set that the wireless connections of interference are mutually formed, interference figure G _i(V _i, E _i) represent in interference relationships;

Step (2), utilizes linear programming to obtain the optimal Decomposition set of optimum link assignment problem

Step (3), for arbitrary interfering nodes v _i, v _i∈ V _i, according to with the optimal Decomposition set that step (2) obtains obtain the weights omega (v of its correspondence _i), wherein e _ijrepresent the limit connecting i and j, connect if existed, then e _ijbe 1, if there is no, then e _ijbeing that 0, F represents handling object, is several data flow f _kthe set formed, λ represents link utilization, C _ijrepresent link capacity, ω (v _e) represent the weight of interfering nodes;

Step (4a), the multiple independent set IS after acquisition decomposition and the mark set { y (I of its correspondence _i);

Step (4b), according to mark Resource selection independent set IS that (4a) obtains, is referred to as I ^*;

Step (5), for arbitrarily for γ _ijassignment, if e _ij∈ I ^*, γ _ijbe 1, represent e _ijbe added into and separate set, otherwise γ _ij=0, represent e _ijbe not added into and separate set;

Step (6), in conjunction with γ obtained in the previous step _ij, upgrade { γ _ijset, obtain new { γ _ij, namely final { γ _ij, for { γ _ijin be 1 element, its represent connection belong to final disaggregation; In conjunction with final { γ _ijand the middle optimal Decomposition set obtained of step (2) obtain represent that whether a kth data flow is through i node and j node;

Step (7a), according to by data flow f _kbe assigned to different paths, obtain a set of paths, the requirement of each paths all demand fulfillment data flow;

Step (7b), for arbitrary data flow f _k, according to probability selection one paths, join set of paths;

Step (8), return path set { P ^kand { γ _ij, terminate computing.

2., according to claim 1 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, available wireless is connected in any limit, put it into interference figure G _i(V _i, E _i) as a node, if any two nodes exist interference, then between these two nodes, add a limit (e ₁, e ₂); For interference figure G _i(V _i, E _i) in node, if there is not limit between two nodes, then show that these two nodes are separate, independent set IS is interference figure G _i(V _i, E _i) a subset, in IS, any two nodes are all independent, in order to set up wireless connections, from interference figure G _i(V _i, E _i) a middle selection independent set IS, assuming that

Wherein, for e _ij∈ E _w, γ _ijrepresent e _ijwhether selected, if choose, be then 1, otherwise be 0, if the connection existence interference that two selected, then cannot meet formula (2), that is, all selected connections are from same independent set IS.

3. according to claim 1 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, optimal Decomposition set in described step (2) method for solving be:

For C _ij

Solve

minλ

Belong to NP-hard problem, use relaxed algorithm by decomposition step, finally utilize linear programming to obtain the optimal Decomposition set of optimum link assignment problem

for binary number, represent data flow f _kwhether through and connection e _ij,

Wherein, δ ₁, δ ₂represent the quantity of the wired and wireless connections in routed path.

4. according to claim 3 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, use relaxed algorithm by after decomposition step, utilize Linear programming thought to obtain a mark set, some points in the middle of corresponding convex combination, namely gather

5. according to claim 3 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, take situation in conjunction with the size of each data flow and every bar limit, and link capacity and link utilization, calculate e _ijweights omega (e _ij).

6. according to claim 3 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, in described step (4a), utilize graph coloring algorithm, by interference figure G _iwith weight set ω (v _i) obtain interference figure G as algorithm input _idecomposition result { y (I _i): I _i∈ I}, the independent set IS after obtaining decomposing and the mark set { y (I of its correspondence _i).

7. according to claim 3 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, in described step (4b), according to mark set y (I ^*) corresponding mark character, select the independent set IS that is optimum.

8. according to claim 1 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, in described step (6), in conjunction with final { γ _ijand the middle optimal Decomposition set obtained of step (2) obtain if be 1, represent that a kth data flow through i node and j node, otherwise, without i node and j node.

9., according to claim 1 based on the dynamic data stream scheduling method of the data center of mixed transmission modes, it is characterized in that, in described step (7a), for if be 1, f _kthrough e _ij, otherwise, without it, final f _kobtain mulitpath, Stochastic choice one transmits.