优化VNF迁移影响的多标准决策方法
183 浏览量
更新于2024-08-27
收藏 2.07MB PDF 举报
"A Multi-Criteria Decision Approach for Minimizing the Influence of VNF Migration"
这篇研究论文探讨了网络功能虚拟化（Network Function Virtualization, NFV）中的一个关键问题——虚拟网络功能（Virtual Network Function, VNF）迁移的影响最小化。在NFV中，网络服务不再依赖特定的硬件设备，而是通过在通用的虚拟化基础设施上部署和管理VNF来实现，这大大提高了服务部署的速度和灵活性。然而，VNF的迁移可能会带来性能下降、服务质量中断以及资源利用率不高的风险。
论文提出了一种多准则决策方法，旨在降低VNF迁移对网络性能和稳定性的负面影响。这一方法考虑了多个因素，包括但不限于：
1. **资源效率**：VNF迁移应尽可能地减少对计算、存储和网络资源的需求，确保资源的高效利用。
2. **服务连续性**：迁移过程中必须保证服务不中断或尽量减少中断时间，以维护用户满意度和业务连续性。
3. **性能影响**：迁移应避免对其他运行中的VNF或网络服务造成性能下降。
4. **成本效益**：考虑到迁移过程可能产生的成本，如能源消耗和管理开销，寻求成本效益最优的解决方案。
5. **VNF共享**：通过合理规划VNF的共享，可以减少迁移次数，进一步减轻对网络的影响。
论文详细阐述了该决策方法的构建过程，包括问题建模、决策准则权重分配、VNF迁移策略的生成和评估等步骤。其中，作者可能采用了如模糊综合评价、层次分析法（Analytic Hierarchy Process, AHP）或其他多准则决策分析工具来处理复杂的决策问题。
文章还可能包含了实验设计和结果分析，通过模拟或真实网络环境验证了所提方法的有效性和优越性。实验结果可能展示了如何通过该决策方法减少VNF迁移带来的影响，优化网络性能，并提高整体系统的稳定性。
此外，论文中可能还讨论了未来的研究方向，比如进一步优化VNF调度算法，增强自适应性和动态调整能力，以及如何更好地应对网络状况的变化和不确定性。
这篇论文为网络功能虚拟化的管理和优化提供了一个实用的多准则决策框架，对于理解和解决VNF迁移问题具有重要的理论和实践意义。
B. Yi, X. Wang and M. Huang et al. / Computer Networks 159 (2019) 51–62 53
considered and solved in this paper. In particular, we assume that
each VNF can be shared by multiple SFCs, such that the migration
of one VNF will inevitably affect all the SFCs traversing it, and our
work intends to minimize such influence, which has not yet been
studied.
Apart from the above related work, there are also many other
works studying the VNF migration problem in clouds. For example,
Cerroni and Callegati [15] proposed a migration model for VNFs, in
which two approaches were proposed for migrating multiple VNFs.
The first one migrated multiple VNFs sequentially, while the sec-
ond one migrated multiple VNFs simultaneously. Despite this, both
of them focused on minimizing the service down time and the to-
tal migration time. Apparently, the sequential migration was easy
to be implemented but inefficient when it required migrating a
large number of VNFs. Meanwhile, the parallel migration was com-
plex but it migrated multiple VNFs efficiently. Thus, the essence of
Cerroni and Callegati [15] was in fact to make the trade-off be-
tween the two schemes. In addition, Hatem et al. [16] specifically
focused on the migration of the virtual content delivery functions.
During the migration process, the flow balance and conservation
were considered in order to minimize the time spent on migration.
Generally, the migration process may cause performance degra-
dation in terms of jitter, packet loss, etc. In order to minimize the
degradation, many effort s have been proposed to decrease the mi-
gration time, thus eliminating performance degradation as much
as possible. One typical example was shown in Ref. [17] which
formulated the VNF migration problem as a mathematical model
and proposed an approach based on linear approximation and fully
polynomial time approximation to solve it. Specifically, Wang et al.
[17] first computed the optimal migration sequence and the re-
quired bandwidth for each migration. Then, based on the sequence
information, it leveraged the calculated bandwidth to guarantee
the migration quality when transmitting the VNF states. Compared
with traditional VM migration algorithms such as Refs. [18] and
[19] , Wang et al. [17] claimed to save the migration time and ser-
vice downtime by up to 40% and 20% respectively. Besides, Nobach
et al. [20] offered VNFs with interfaces which were used to an-
nounce the changing states for incoming packets. In this way, the
VNF state synchronization could be obtained efficiently with a low
cost, and such synchronization mechanism enabled an efficient and
seamless VNF state migration.
Nevertheless, we can discover that these related work actually
focused on the influence (e.g., service down time) caused by the
migration process. However, on one hand, VNFs have much smaller
size than VMs and we may only need to migrate the stateful infor-
mation of VNFs, such that the migration process won’t last long.
On the other hand, due to the VNF sharing among different SFCs,
the migration of VNF will inevitably affect multiple SFCs. Therefore,
instead of focusing on the migration process, we prefer to explore
the potential of minimizing the influence generated on all the SFCs
after VNF migration, which is also meaningful and unexplored yet.
3. System model and problem description
In this section, we build a mathematical model for VNF migra-
tion and all the notations used are shown in Table 1 .
3.1. Physical network
The physical network is abstracted as a graph denoted by G
p
( N
p
,
E
p
) where N
p
is the set of physical nodes and E
p
is the set of
physical links. In particular, N
p
= { n
p
i
| i ∈ [1 , | N
p
| ] } and E
p
= { e
p
j
| j ∈
[1 , | E
p
| ] } , where n
p
i
is the i th physical node while e
p
j
is the j th
physical link.
For each physical node n
p
i
, its available resources are denoted
by R
n
p
i
, since we consider all the resources on physical nodes as
one kind. For each physical link e
p
j
, its available bandwidth re-
source is denoted by R
e
p
j
and its delay is denoted by D
e
p
j
respec-
tively.
3.2. VNF forwarding graph
The VNF FG can actually be regarded as a kind of virtual net-
work which is composed of VNFs. In this virtual network, each VNF
can be regarded as a virtual node while the connection between
any two adjacent VNFs can be regarded as a virtual link.
Therefore, this VNF FG can also be abstracted as a graph de-
noted by G
v
( N
v
, E
v
) where N
v
is the set of virtual nodes (i.e., VNFs)
and E
v
is the set of virtual links. Similarly, N
v
= { n
v
i
| i
∈ [1 , | N
v
| ] }
and E
v
= { e
v
j
| j
∈ [1 , | E
v
| ] } , where n
v
i
is the i
th virtual node while
e
v
j
is the j
th virtual link.
In particular, we assume that these VNFs are already deployed
in the physical network. Thus, there is a mapping relationship be-
tween the physical network and VNF FG. Such relationship is ex-
pressed by the following two variables, where X
n
p
i
n
v
i
indicates the
mapping between virtual nodes and physical nodes and Y
e
p
j
e
v
j
indi-
cates the mapping between virtual links and physical links.
X
n
p
i
n
v
i
=
1 , If n
v
i
is mapped on n
p
i
.
0 , Otherwise.
(1)
Y
e
p
j
e
v
j
=
1 , If e
v
j
is mapped on e
p
j
.
0 , Otherwise.
(2)
Note that, the processing time of each VNF on serving different
SFCs should be considered. Due to the situation that each VNF is
shared by multiple SFCs and it can only process the traffic of one
SFC at the same time, the M/M/1 queueing model [21] is adopted.
Given any VNF instance n
v
i
in VNF FG, its service time is denoted
by r
v
i
following exponential distribution.
3.3. Service function chain
There are many SFCs in the network and we denote them as
follows:
S = { s
μ
| μ ∈ [1 , | S| ] } , (3)
where s
μ
indicates the μth SFC.
For any SFC s
μ
, it requires a set of VNFs which are denoted as
follows:
F
μ
= { f
ν
μ
| μ ∈ [1 , | S| ] , ν ∈ [1 , | F
μ
| ] } , (4)
where f
ν
μ
is the νth VNF required by s
μ
.
For any f
ν
μ
, it has a resource requirement denoted by R
ν
μ
, and
the bandwidth required by s
μ
to connect all these VNFs is denoted
by B
μ
. In addition, s
μ
allows for a maximum delay denoted by D
μ
.
In particular, all these VNFs required by SFCs also have a map-
ping relationship with those formulated in VNF FG. Thus, the VNF
supplement relationship is described by the following variable.
Z
n
v
i
f
ν
μ
=
1 , If f
ν
μ
is supplied by n
v
i
.
0 , Otherwise.
(5)
It is worth noting that different SFCs can share the same VNF
as long as the available resources are enough. In addition, in or-
der to chain these required VNFs, we should build the connections
between any two adjacent VNFs as follows:
Z
e
v
j
f
ν
μ
=
1 , If virtual link e
v
j
is used to connect f
ν
μ
and f
ν+1
μ
.
0 , Otherwise.
(6)
where μ ∈ [1, | S |] and ν ∈ [1, | F
μ
|).
剩余11页未读，继续阅读
2019-08-22 上传
2021-02-08 上传
2017-09-30 上传
2021-02-22 上传
2011-07-02 上传
2021-02-08 上传
2021-06-05 上传
2021-05-28 上传
2021-05-24 上传
weixin_38642897
- 粉丝: 3
- 资源: 895
我的内容管理
展开
最新资源
- 基于Python和Opencv的车牌识别系统实现
- 我的代码小部件库：统计、MySQL操作与树结构功能
- React初学者入门指南：快速构建并部署你的第一个应用
- Oddish:夜潜CSGO皮肤，智能爬虫技术解析
- 利用REST HaProxy实现haproxy.cfg配置的HTTP接口化
- LeetCode用例构造实践：CMake和GoogleTest的应用
- 快速搭建vulhub靶场：简化docker-compose与vulhub-master下载
- 天秤座术语表：glossariolibras项目安装与使用指南
- 从Vercel到Firebase的全栈Amazon克隆项目指南
- ANU PK大楼Studio 1的3D声效和Ambisonic技术体验
- C#实现的鼠标事件功能演示
- 掌握DP-10：LeetCode超级掉蛋与爆破气球
- C与SDL开发的游戏如何编译至WebAssembly平台
- CastorDOC开源应用程序：文档管理功能与Alfresco集成
- LeetCode用例构造与计算机科学基础：数据结构与设计模式
- 通过travis-nightly-builder实现自动化API与Rake任务构建
