关系图驱动的SDN高效增量更新策略

需积分: 5 178 浏览量更新于2024-08-26 收藏 1.64MB PDF 举报

在软件定义网络（Software-defined networking, SDN）的背景下，网络配置一致性在更新过程中对网络稳定性至关重要。为了克服这一挑战，本文提出了一种名为"快速增量一致更新方案"（Fast and Incremental Consistent Update Scheme, FICUS）的方法，其核心是建立在关系图（Relation Graph, RG）基础上的创新架构。首先，文章指出传统的SDN配置更新可能存在效率低下、一致性难以保证的问题。FICUS针对这些问题，通过分析更新操作之间的相互关系，构建了一个全新的关系图结构。这种结构有助于捕捉和理解网络设备间的复杂依赖关系，如路由、策略和连接等，从而有效地组织和规划更新操作。在关系图模型中，每个节点代表一个网络组件或配置项，边则表示这些组件之间的关联性。通过分析关系图的拓扑，FICUS能够确定更新操作的最佳执行顺序，确保在进行一次增量更新时，不会导致当前网络配置的不一致状态。这不仅减少了不必要的网络中断时间，还降低了因并发更新引发的冲突概率。为了实现快速和一致的更新，FICUS采用了分阶段的更新策略。首先，它会进行预处理阶段，识别出可以并行执行的更新任务，并根据关系图中的路径优先级确定更新路径。然后，在实际执行阶段，按照确定的顺序逐步执行更新操作，同时通过版本控制机制跟踪每个配置的状态，确保在整个过程中保持一致性。此外，文中还提到了该方案在接收请求（Received 14 September 2016）、修订（Revised 4 March 2017）和接受（Accepted 5 May 2017）过程中的关键步骤，以及最终的在线可用日期（Available online 11 May 2017）。关键词包括SDN配置更新、更新一致性、关系图等，这些都是论文的核心关注点。基于关系图的SDN快速增量更新方案是一项创新性的技术，它通过优化配置更新流程，提高了网络的稳定性和管理效率。这对于大规模SDN网络的部署和维护具有显著的实际价值，为后续的研究和工程实践提供了重要的理论支撑。

Q. Li et al. / Computer Networks 125 (2017) 41–52 43

alive. Since an OpenFlow switch is very simple and has little intel-

ligence, only the switch rules with a timeout mechanism can be

achieved. However, completing the update using this mechanism

is time consuming.

To reduce the cost of TCAM, an incremental update algorithm

[13] that splits the update into several rounds has been proposed.

Each round consists of the following: 1) choose a subset of ﬂows

that are moved from the old conﬁguration to the new conﬁgura-

tion; 2) ﬁnd new rules to be installed according to the ﬂows; 3)

install rules using the two-phase commit mechanism; and 4) ﬁnd

old rules that are not used by any ﬂows and delete them. S. Viss-

icchio et al. [25] present a method to complete anomaly free up-

dates in hybrid networks. The trade-off between the strength of

the consistency property and dependencies among rules is argued

in [26] . The controller generates an updated DAG according to the

consistent property and topology, which is complex and requires a

number of computations.

Our mechanism is developed from the model proposed in

[7,27] . These modes only focus on a static conﬁguration and lack

a formal description and deﬁnition. A network formalism has been

explored in the veriﬁcation of network functions [28] . We ex-

tended the network semantics to include switches, updates and

consistent priorities; hence, our mechanism is also easy to be ex-

tended to ﬂow-level consistency properties.

3. The update model

In this section, we present a formal model to clearly describe

the network update process. In our model, we abstract the packet

as a point and the rule as a hypercube in the match space. Accord-

ing to the behaviors of packets in the data plane, we construct the

RG for the update operations. With the help of the RG, we can de-

termine the appropriate order to install the update rules. To clearly

describe the update process, we ﬁrst deﬁne some terminology.

Rule, r . A rule in an SDN switch contains match ﬁeld, action,

priority and other information (non-signiﬁcant in this paper). In

our discussion, we deﬁne a rule as a tuple r = (m, a, pri ) . The ﬁrst

element m denotes the match ﬁeld that uniquely identiﬁes the

rule. The second element a is the action of the rule. The last el-

ement pri is an integer that represents the priority of the rule. Let

rs be the default rule set in our discussion. We use ms to denote

the set of match ﬁelds, ms

to denote the match ﬁeld set of rules

with priorities that are less than or equal to i , and

to denote

the match ﬁeld set of rules with priorities that are higher than i .

We deﬁne the real match space of a rule ( m, a, pri ) as re (m ) =

m − m



pri

. If there are several different rules that match the

same packet p, we deﬁne a new relation ( p  r

) to represent the

matching ( r

) rule with the highest priority.

Topology, T . We deﬁne the network position lp as a two-tuple

( s, p ), where s is the switch and p is the ingress port of switch

s . Let LP be the set of all the possible network positions from the

perspective of the data plane (the controller is not included). We

deﬁne the topology function as follows: T (lp) = lp



indicates that

lp and lp



are directly connected.

Conﬁguration, C . In SDN, the conﬁguration C is the collection

of rules that are installed in the switches. Thus, the conﬁguration

can be described as a set {( lp

, rule

)| lp

∈ LP, rule

∈ rs }.

Network, N . The network N is also deﬁned as a tuple ( C, T ), in

which topology T denotes the position and connection information

between network devices and conﬁguration C denotes the forward-

ing behaviors in the data plane.

Network Update Process. In SDN, operators enforce a cen-

tralized management through the controller, including changing

the network conﬁgurations, obtaining the statistics and so forth.

Changing the network conﬁgurations corresponds to installing and

deleting rules on the switches. Since operations such as collecting

the statistics do not affect the forwarding behavior, we do not fo-

cus on such types of operations in this paper.

An update operation must clearly have a location where the

rule is installed or deleted. Hence, we deﬁne update operation u

as ( lp, r ), which means that the new rule r should be installed at

network position lp . We use us to denote an update sequence. For

each packet pkt , there is a trace t to record its forwarding behavior

in the data plane. Trace t is deﬁned as an ordered list (i.e., [( lp

), ( lp

, r

), , ( lp

, r

)]), where lp

is the i th network position

along the path of packet pkt and r

is the corresponding rule for

the packet at lp

. We use the expression lp

∈ t to indicate that lp

appears in the trace t . A similar expression is also suitable for r

∈ t and u

∈ C

. According to the deﬁnition of update operation, a

trace can also be described as an ordered list of update operations,

i.e., [ u

, u

, , u

]. The trace of a packet describes its forwarding

behaviors in the data plane; thus, the controller can easily obtain

the current traces and infer the new traces after update operations.

In our paper, the network update process can be deﬁned as

follows:

if C



= ov er r ide (C, us )

then N



−→ N



where N = (C, T ) , N



= (C



, T )

The function override ( C, us ) means that for each two-tuple u =

(lp, r) ∈ us, if u ∈ C , add u to C



; if u ∈ C , do nothing. We set a

higher priority for the updated (new) rules compared with the old

rules. The expression N



−→ N



means that under the conﬁguration

of C



, the network N changes to N



Consistency property. In SDN, the update message is forwarded

to the switch from the controller, and then the switch updates its

ﬂow table accordingly. The switch is only responsible for forward-

ing packets. The consistency property should be guaranteed by the

controller during the update process since forwarding faults may

occur among the chaos of old and new conﬁgurations.

For different networks, there are different consistency proper-

ties. For instance, packet-level consistency means that all packets

should follow the same policy, i.e., they behave according to ei-

ther the new conﬁgurations or the old one but not according to

mixed conﬁgurations. Conversely, ﬂow-level consistency maintains

the consistency property at a more coarse granularity. In our paper,

we focus on the packet-level consistency property.

Generally, the connectivity should ﬁrst be guaranteed in the

network. Hence, we propose the path dependency relation among

the update operations, which is used to ensure that the switch can

correctly forward packets during the update. Since this relation can

guarantee loop-free and black-hole-free properties during the up-

date, we then propose the path rejection relation among update

operations to guarantee the packet-level consistency. In practice,

for different consistency properties, different relations can be pro-

vided. Therefore, an RG can be generated according to these re-

lations, and the proper update order will be discovered. We will

present the details of our RG model in the next section.

4. Fast incremental consistent update

In this section, we describe the details of FICUS. In SDN, the

controller translates the policy into the conﬁgurations and then in-

stalls them into the switches. The key concept of our mechanism

is analyzing the relationship between the update operations and

determining the proper order to install them incrementally. In our

scheme, we guarantee the packet-level consistency property, which

means that for a certain packet, it will only follow one conﬁgura-

tion, either old or new. This property can ensure the correctness of

forwarding behaviors during the update process.

剩余11页未读，继续阅读

weixin_38708707

粉丝: 5

关系图驱动的SDN高效增量更新策略

FICUS：SDN中的快速增量一致更新技术

SDN策略更新：基于路径与反馈的解决方案

基于SDN的虚拟机迁移网络优化方案研究

FICUS：基于关系图的SDN中的快速增量一致更新

通过增量SDN部署缓解链路泛洪攻击

SDN中基于链路状态感知的路由优化模型

SDNPS：软件定义网络中基于负载平衡的基于主题的发布/订阅系统

思科设备备份模式：增量备份与差异备份的终极区别与应用指南

云服务与VMware整合方案：构建高效混合云架构

【高可用性架构设计】：基于IB Specification Vol 1 Release 1.3的架构构建

最新资源