JP6178743B2 - Traffic flow allocation method and apparatus - Google Patents

Description

  The present invention relates to a traffic flow assignment method and apparatus for assigning a route on a network to a traffic flow group to which attributes such as arrival / departure nodes are given in advance, and in particular, each slice (virtual virtual network) of an SDN (Software-Defined Network). The present invention relates to a traffic flow allocating method and apparatus for realizing sharing of a link bandwidth by sharing a packet transfer queue between networks with a small amount of computing resources.

  Non-Patent Documents 1 and 2 disclose SDN as a technology for creating a virtual network by software. In SDN, since a number of completely independent virtual networks called “slices” can be constructed on one physical network, a unique network corresponding to a user's request can be constructed for each slice.

  Patent Document 1 discloses a traffic flow allocation method and apparatus for promoting sharing of a packet transfer queue and a link band between slices in an SDN to minimize a packet transfer scheduling cost.

  In Patent Document 2, a plurality of virtual network candidates having a tree topology are configured in advance for each slice of the SDN, and an integer programming model is solved to minimize the packet transfer scheduling cost. A traffic flow allocation method and apparatus for selecting a combination of the above is disclosed.

  In Patent Literature 3, a plurality of virtual network candidates are configured by sequentially allocating traffic flows to each slice of the SDN using a link cost that reflects an increase in the inter-slice scheduling cost. A technique for selecting a combination of virtual network candidates that minimizes a packet transfer scheduling cost between slices by solving a legal model is disclosed.

Japanese Patent Application No. 2013-262879 Japanese Patent Application No. 2013-269103 Japanese Patent Application No. 2014-050330

"Examination of Integrated State Collection and Analysis Mechanism for Control Plane Application Development and Operation Monitoring in Software Defined Network" IEICE Technical Report. NS, Network Systems 111 (408), 127-132, 2012-01- 19 "OpenFlow / SDN and Network Virtualization" IEICE Technical Report, vol. 112, no. 230, IN2012-96, pp. 115-119, October 2012, Shigenori Shindo

  Each slice of SDN can share each link of the physical network. However, if a unique link bandwidth is set for each slice, the link bandwidth cannot be used effectively. For example, if a 1G bandwidth is fixedly allocated to slices A and B for a physical link with a bandwidth of 2G, for example, even if slice A has a large traffic volume and slice B has a small traffic volume, slice A Cannot allocate the surplus bandwidth of slice B.

  On the other hand, if the bandwidth is allocated so that slices A and B share one packet transfer queue for each physical link, each slice A and B can allocate the surplus bandwidth to other slices. The link bandwidth can be effectively used without degrading the service quality of the slice. However, until now, it has not been studied that each slice of an SDN shares a link bandwidth by sharing one packet transfer queue for each physical link.

  In response to such a technical problem, the inventors of the present invention input the network topology information and traffic flow information as input, and assign traffic flow allocation that minimizes the sum of scheduling costs for packet transfer of all links to an integer plan. Calculated as a solution of the legal model, and for each ring of the network, among the slices with one priority class of traffic flow passing through the link, the link bandwidth is set between slices having the same priority class. A method and apparatus for calculating traffic flow allocation under shared conditions was invented and a patent application was filed (Patent Document 1).

  However, the direct solving method of the integer programming model disclosed in Patent Document 1 uses a large amount of computing resources, and thus has a technical problem that it is difficult to apply to a large-scale SDN.

  In response to such a technical problem, the inventors of the present invention further select a tree topology candidate that most reduces the packet transfer scheduling cost from a plurality of tree topology candidates given in advance in each slice. Only by solving the integer programming model, a traffic flow allocation method and apparatus that do not require a large amount of computational resources were invented and a patent application was filed (Patent Document 2).

  However, a virtual network having a tree topology in each slice does not always reduce the scheduling cost of packet transfer.

  In response to such a technical problem, the inventors of the present invention further create a plurality of virtual network candidates that satisfy all flow conditions for each slice and have different paths, and select virtual network candidates selected from each slice. Invented a traffic flow allocation method and apparatus for calculating a combination that minimizes the sum of scheduling costs related to packet transfer of each link through which traffic flows pass as combinations of the number of slices, as a solution of an integer programming model. Applied (Patent Document 3).

  In Patent Document 3, an integer programming model solution is obtained only for selecting a combination of virtual network candidates that reduce the packet transfer scheduling cost most from a plurality of virtual network candidates that are created for each slice and have different paths. Since it only has to be done, it does not require a large amount of computing resources, and can be applied to large-scale SDN.

  However, according to Patent Document 3, although the computational resources required for solving the integer programming model can be greatly reduced as compared with the prior art, if the number of slices increases, the corresponding computational resources are required. There was a technical problem that it was difficult to apply.

  The object of the present invention is to solve the above technical problem and promote the sharing of the packet transfer queue between the slices and the accompanying link bandwidth within the range that does not affect the packet transfer quality in each slice of the SDN. Thus, it is an object of the present invention to provide a traffic flow allocation method and apparatus for realizing traffic flow allocation that minimizes the packet transfer scheduling cost between slices with a small amount of calculation.

  In order to achieve the above object, the present invention has the following configuration in traffic flow allocation for calculating traffic flow allocation for realizing link bandwidth sharing by sharing packet transfer queues between slices of SDN: There is a feature.

  (1) The traffic flow allocation apparatus of the present invention determines the cost of each link based on the allocation of traffic flow paths already completed for each traffic flow path allocation. A means for updating the expected value of the inter-slice scheduling cost when it is assumed to pass, and updating the link cost for each slice while sequentially assigning a minimum cost path to all traffic flows, And means for repeating the slice.

  (2) In the traffic flow allocation method of the present invention, the cost of each link is set to an increase amount of the expected value of inter-slice scheduling cost when it is assumed that the traffic flow to be calculated by the route calculation passes the link to be cost set; Already completed traffic flow for each traffic flow path assignment, and the procedure of assigning the least cost path sequentially to each traffic flow for each slice and repeating this for all traffic flows. A procedure for updating based on route allocation and a procedure for repeating sequential allocation of minimum cost routes and updating of link costs for all slices are provided.

According to the present invention, the following effects are achieved.
(1) In each slice of the SDN, traffic flow allocation that promotes sharing of a packet transfer queue and a link band between slices can be realized, so that resource utilization efficiency can be improved.

  (2) Since a traffic flow path that minimizes the sequential increase in expected packet transfer scheduling cost between slices is calculated, traffic flow allocation that reduces the packet transfer scheduling cost between slices can be realized.

  (3) Since the minimum cost route calculation is only executed twice for each traffic flow, traffic flow allocation to reduce the packet transfer scheduling cost between slices can be performed with a small amount of calculation even in a large-scale SDN with a large number of slices. realizable.

It is the figure which showed the typical example of sharing of a link band by sharing of the packet transfer queue between slices by this invention. It is the figure which showed the example which is not suitable for sharing of a link band by sharing of a packet transfer queue because the priority class is set to the traffic flow. FIG. 11 is a diagram (part 1) illustrating an example suitable for sharing a link bandwidth by sharing a packet transfer queue even when a priority class is set for a traffic flow; FIG. 10 is a diagram (part 2) illustrating an example suitable for sharing a link bandwidth by sharing a packet transfer queue even when a priority class is set for a traffic flow; It is a figure for demonstrating the relationship between the number of slices which share a link, the packet transfer queue set to each slice, and scheduling cost. It is the functional block diagram which showed the structure of the principal part of the traffic flow allocation apparatus which concerns on one Embodiment of this invention. It is the flowchart which showed the procedure which calculates all the traffic flow paths which satisfy | fill a flow condition for every slice. It is the figure which showed the other example of sharing of a link band by sharing of the packet transfer queue between slices by this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. First, link bandwidth sharing by sharing packet transfer queues according to the present invention will be described, and then a traffic flow allocation method capable of minimizing scheduling cost by sharing link bandwidth will be described in detail.

  The scheduling cost noticed by the present invention is a processing cost required for each node to transfer a packet of each traffic flow with a priority according to its priority class. The scheduling cost depends on the number of slices sharing the link bandwidth and the number of packet transfer queues. Dependent.

  In the present invention, typically, as shown in FIG. 1, the packet transfer queues provided in the same physical link in a plurality of slices (here, two slices # 1 and # 2) are combined into one. By sharing a single packet transfer queue for multiple slices # 1 and # 2, the total of the link bandwidth allocated to each of slices # 1 and # 2 for that link is shared by slices # 1 and # 2 Traffic flow allocation is realized.

  On the other hand, in a slice in which a plurality of priority classes are provided in a traffic flow, band allocation that is relatively superior or inferior is generally performed according to the priority class. However, since such bandwidth allocation is based on contract service charges, etc., even if packet transfer opportunities with lower priority classes are lost as a result of the increase in packet transfers with higher priority classes, the quality of service is reduced. There is not much unfairness inside.

  However, as shown in FIG. 2, when the link bandwidth is shared by slices (# 3, # 4) in which a plurality of priority classes are provided in the traffic flow, one slice (here, slice # 4) If the traffic flow increases and the high-priority traffic flow increases, the bandwidth that should be allocated to the low-priority traffic flow of the other slice (here, slice # 3) is eroded, and the service quality is degraded. So unfairness arises.

  Therefore, in the present invention, the slices to be shared of the link bandwidth by sharing the packet transfer queue are limited to slices in which only a single and the same priority class traffic flow flows in each link.

  In addition, as shown in FIG. 1, the combination of slices includes not only the slices having a unique priority class as shown in FIG. 1, but also traffic flows having different priority classes as slices as shown in FIGS. However, in the link unit, slices in which only the high priority traffic flow flows (FIG. 3) and slices in which only the low priority traffic flow flows (FIG. 4) are also the same in the slices in which the priority class is unique. The slice is a link band sharing target.

  1 to 4, for the convenience of explanation, the packet transfer queue and link bandwidth sharing between two slices has been described as an example. However, as shown in FIG. The same applies to links shared by # 6. That is, the slices # 1 and # 2 are not shared because there are a plurality of priority classes of traffic flows. Slices # 3 and # 4 are shared because only the high-priority traffic flow passes through them. Slices # 5 and # 6 are shared because only the low-priority traffic flow passes.

  The present invention considers the realization of traffic flow allocation that promotes sharing of packet transfer queues and link bandwidths between slices and minimizes packet transfer scheduling costs between slices in the entire network. Here, the packet transfer scheduling cost between slices to be minimized is the sum of the scheduling costs between slices in each link. The inter-slice scheduling cost in each link is represented by the number of virtual machines that process packet transfer control in each slice, and is simply proportional to the number of slices after link bandwidth sharing. Therefore, in the example of FIG. 5, the number of virtual machines is 4, and the scheduling cost between slices is proportional to 4.

  FIG. 6 is a functional block diagram showing the configuration of the main part of the traffic flow allocation apparatus 1 according to one embodiment of the present invention. Such a traffic flow allocation apparatus 1 can be configured by installing an application (program) for realizing each function described later on a general-purpose computer or server. Alternatively, a part of the application may be configured as a dedicated machine or a single-function machine in which hardware or ROM is implemented.

  The input interface (I / F) 10 includes a topology information receiving unit 10a that receives and stores input of network topology information, and a traffic flow information receiving unit 10b that receives and stores input of traffic flow information. The minimum hop number route calculation unit 20 calculates the minimum hop number route of all traffic flows, and further calculates the average number of via hops.

  The traffic flow route calculation unit 30 includes a link cost setting unit 30a, and, as will be described in detail later, for all traffic flows, a traffic flow route that minimizes an increase in the expected packet transfer scheduling cost between slices. Calculate sequentially. In order to calculate the expected value of the packet transfer scheduling cost between slices, the value of the average number of hops in the traffic flow is necessary. Use this instead.

  The link cost setting unit 30a assumes that the cost of each link passes through the cost setting target link based on the already assigned traffic flow path allocation for each traffic flow path allocation. Update to the expected increase in inter-slice scheduling cost. The sequentially calculated traffic flow route is output as a traffic flow allocation result that minimizes the scheduling cost of packet transfer.

  FIG. 7 is a flowchart showing a procedure in which the traffic flow path calculation unit 30 sequentially calculates a traffic flow path for each slice. All the slices are selected in order, and the traffic accommodated in the selected slice is shown. The flow path is calculated sequentially one by one. Prior to route calculation of each traffic flow, the cost of each link is set to an increase in the expected value of inter-slice scheduling cost when the traffic flow subject to route calculation passes through the link. The cost of each link is set in consideration of the allocation result of traffic flows accommodated in other slices that have already been clarified.

  That is, in step S101, a list of slices to be shared with the link band is created. In step S102, one slice (target slice) of interest is extracted from the slice list.

  In step S103, the flow list of the target slice is created based on the traffic flow information. In step S104, one traffic flow is selected at random from the flow list to be the current traffic flow of interest, that is, the route calculation target traffic flow.

  In step S105, for each link, the link cost that the traffic flow passing through the link bears is set and updated. In the present embodiment, as will be described in detail later with reference to equations (1) to (5), the cost of each link is calculated based on the traffic flow route allocation that has already been completed. It is updated to the increase amount of the expected value of the inter-slice scheduling cost when it is assumed that the cost setting target link is passed.

  In step S106, a minimum cost route of the traffic flow of interest is calculated based on the set current link cost. In step S107, it is determined whether or not the allocation of the minimum cost path has been completed for all traffic flows registered in the flow list. If not completed, the process returns to step S104, and the traffic flow selection, link cost update, and minimum cost path assignment are repeated while switching the traffic flow of interest within the same slice.

  Thereafter, when it is determined that the processing related to all traffic flows registered in the flow list of the target slice has been completed, the process proceeds to step S108, and all the minimum cost routes calculated in the current target slice are registered as traffic flow routes. To do.

  In step S109, it is determined whether or not the calculation of the traffic flow path for all slices is completed and the slice list becomes empty. If not completed, the process returns to step S102, and the above processes are repeated while switching the target slice, and a virtual network is constructed for each slice.

  Next, the link cost setting method in step S105 will be described in detail. In the present embodiment, each link cost Cost in the virtual network construction of each target slice is an increase in the expected value of inter-slice scheduling cost (0 to 1) when it is assumed that the current traffic flow passes through the cost setting target link. And takes one of the following three values depending on the attribute of the link. Note that the proportional coefficient of the inter-slice scheduling cost with respect to the number of slices is assumed to be 1.0.

Case 1 Link with no passing traffic flow for the slice in which the traffic flow of interest is accommodated: Cost = LC0
Case 2 Regarding a slice in which the traffic flow of interest is accommodated, a link through which only a traffic flow belonging to one priority class different from the traffic flow of interest passes: Cost = 1.0-LC0
Case 3 Other links: Cost = ε (ε is a sufficiently small value)

  That is, regarding the slice in which the traffic flow of interest is accommodated, when the traffic flow of interest passes first from the state where no traffic flow passes through the link whose cost is to be set (case 1), the traffic flow that has already been accommodated. The number of slices in which only the traffic flow belonging to the same priority class as the traffic flow of interest passes through the link among the slices for which the route has been determined is N, including the slice that accommodates the traffic flow of interest. Let S be the number of slices for which the traffic flow route to be determined has not been determined.

  Among the slices for which the traffic flow route to be accommodated other than the current slice is not determined, if the i-slice and the link can be expected to share bandwidth, the increase in the inter-slice scheduling cost for the link is 1 If / (i + 1) has already been determined that the bandwidth can be shared for the N slices and the link, the increase in the inter-slice scheduling cost in the link can be expressed by the following equation (1).

  For example, if there is not yet a slice in which only traffic flows belonging to the same priority class as the traffic flow of interest pass through the link among slices for which traffic flow paths to be accommodated have already been determined (N = 0), The increase in the scheduling cost between slices in the link is 1 / (i + 1).

  Furthermore, when bandwidth sharing of the link with the remaining slices other than the current target slice for which a traffic flow path is to be obtained cannot be expected at all (i = 0), the increase in the inter-slice scheduling cost in the link is It becomes 1 as it is.

  Note that a slice that can share the link band is a slice in which only one or more traffic flows belonging to the same priority class as the traffic flow of interest pass through the link.

  If the number of combinations for selecting n slices from m slices is mCn and the probability that only one or more traffic flows belonging to the same priority class as the traffic flow of interest passes through the link is P The probability that the link bandwidth can be shared with i (= 0 to S-1) slices of S-1 slices other than the target slice that accommodates the traffic flow is expressed by the following equation (2). I can express.

  Therefore, the amount of increase in inter-slice scheduling cost when the link bandwidth can be shared with N slices based on the allocation result of traffic flows accommodated in other slices that have already been clarified is the calculation result of the above equation (2) Can be averaged with respect to i (= 0 to S-1), and can be expressed by the following equation (3).

  On the other hand, regarding the probability P, if the average number of hops through traffic flows is H and the number of links is L, the probability that each traffic flow passes through a link whose cost is to be set is given by (H / L). However, in the present embodiment, the average via hop count H is substituted with the average via hop count of the minimum hop count route of all traffic flows obtained previously.

  If each slice contains Fs traffic flows, the probability that one or more i (= 1 to Fs) traffic flows out of the Fs traffic flows will pass through the link. And can be expressed by the following equation (4)

  In addition, i (= 1 to Fs) traffic flows pass through the link, and all i (= 1 to Fs) traffic flows that pass through the link have the same priority class as the traffic flow of interest. The probability of belonging can be expressed by the following equation (5), where K is the number of priority classes. However, it is assumed that the number of traffic flows belonging to each priority class is equal.

  From the above, in each slice, the probability P that only one or more traffic flows belonging to the same priority class as the traffic flow of interest passes through the link is obtained by taking the sum of i (= 1 to Fs) and calculating the following equation ( It can be expressed in 6).

  In case 2 described above, if the traffic flow of interest passes through the link, at that time, the slice becomes an independent slice that cannot share the link bandwidth with other slices. Therefore, the increase amount of the inter-slice scheduling cost expectation value from the state where there is no traffic flow passing through the link is 1.0, and LC0 represents the cost immediately before the link.

  In case 3, the expected value of inter-slice scheduling cost does not increase even if the traffic flow of interest passes through the link. Therefore, ε is a sufficiently small real number, for example, a number smaller than the reciprocal of the total link number L of the network.

  By calculating the minimum cost route under this link cost, it is possible to calculate a route that minimizes the increase in the expected value of the packet transfer scheduling cost between slices. When the calculation of all the traffic flows belonging to the slice is completed, a virtual network satisfying all the traffic flows can be obtained by combining all the traffic flows.

  As described above, in the traffic flow route calculation procedure in this embodiment, each traffic flow is assigned to a route that minimizes the sequential increase in the expected packet transfer scheduling cost between slices. Traffic flow allocation can be realized to reduce the packet transfer scheduling cost.

  According to the present embodiment, in each slice of the SDN, traffic flow allocation that promotes sharing of a packet transfer queue and a link band between slices can be realized, so that resource utilization efficiency can be improved.

  In addition, according to the present embodiment, a traffic flow path that minimizes the sequential increase in the expected value of the packet transfer scheduling cost between slices is calculated, so traffic flow allocation that reduces the packet transfer scheduling cost between slices is performed. realizable.

  Further, according to the present embodiment, the minimum cost route calculation for calculating the minimum hop count route for each traffic flow and the minimum cost route calculation for determining the assigned route of the traffic flow are combined twice. Since only execution is performed, traffic flow allocation for reducing the packet transfer scheduling cost between slices can be realized with a small amount of calculation even in a large-scale SDN having a large number of slices.

  In the above-described embodiment, the slice set that is the target of sharing the link band is the same and the same slice of the priority class of the traffic flow passing through the link, and the slices through which a plurality of traffic flows of different priority classes pass. Was described as being out of scope for sharing.

  However, even if the slices pass through a plurality of traffic flows with different priority classes, as shown in FIG. 8, if the link band is fixedly assigned to each priority class in the slice, the corresponding priority is set. By summarizing and sharing the packet transfer queue between classes, the sum of the link bandwidths may be shared for each priority class. At this time, the bandwidth allocated to the corresponding priority class of each slice does not necessarily have to be the same.

  In this way, packets with the same priority class are treated fairly, and even if the high-priority traffic flow of one slice increases, the bandwidth to be allocated to the low-priority traffic flow of the other slice is secured. . Therefore, link bandwidth utilization efficiency can be improved without unfairly degrading service quality.

  DESCRIPTION OF SYMBOLS 1 ... Traffic flow allocation apparatus, 10 ... Input interface, 10a ... Topology information reception part, 10b ... Traffic flow information reception part, 20 ... Minimum hop number path | route calculation part, 30 ... Traffic flow path | route calculation part, 30a ... Link cost setting part

Claims (3)

In a traffic flow allocation device that calculates traffic flow allocation that realizes link bandwidth sharing by sharing packet transfer queues between each slice of SDN,
Expected value of inter-slice scheduling cost when it is assumed that the traffic of the link calculation target traffic flow passes through the cost setting target link based on the traffic flow route assignment already completed for each traffic flow route assignment. Link cost setting means for updating to an increase amount of
A traffic flow comprising: a traffic flow route calculation unit that sequentially assigns a minimum cost route to all traffic flows while updating the link cost for each slice and repeats this for all slices. Allocation device. A means for calculating an average number of hops through the traffic flow;
The traffic flow allocation apparatus according to claim 1, wherein the inter-slice scheduling cost expectation value is obtained based on the average number of via hops. In a traffic flow allocation method for calculating traffic flow allocation that realizes sharing of a link bandwidth by sharing a packet transfer queue between slices of SDN,
A procedure for setting the cost of each link to an increase in the expected value of the inter-slice scheduling cost when the path calculation target traffic flow is assumed to pass through the cost setting target link;
For each slice, assign a minimum cost path sequentially to each traffic flow and repeat this for all traffic flows;
Updating the cost of each link based on traffic flow path assignments already completed for each traffic flow path assignment;
A traffic flow allocation method comprising: sequentially assigning the minimum cost path and repeating the update of link cost for all slices.