节能与延迟权衡：移动边缘计算中能源感知的负载优化策略
43 浏览量
更新于2024-08-31
收藏 1.14MB PDF 举报
本文探讨了"能源与延迟权衡在移动边缘计算网络中的能源感知卸载策略"（Energy-Latency Tradeoff for Energy-Aware Offloading in Mobile Edge Computing Networks）。这项研究发表在2018年8月的《计算机互联网期刊》(IEEE INTERNET OF THINGS JOURNAL)第5卷第4期，是该领域具有高引用率的1%文章。随着移动边缘计算（MEC）的发展，它将计算能力带到移动网络的边缘，靠近智能移动设备（SMDs），从而节省了本地计算的能量消耗。然而，这可能导致网络负载增加和传输延迟的提高。
研究的核心关注点在于理解在有限的能源和敏感的延迟条件下，通信和计算资源分配之间的平衡。作者提出了一个能源感知的卸载方案，旨在同时优化这两方面的资源。该方案考虑了单个和多细胞MEC网络场景，引入了智能设备电池剩余能量作为衡量能源消耗和延迟权重的因素。通过混合整数非线性优化方法，论文解决的是计算卸载任务和资源分配的问题。
具体来说，论文可能会分析不同的通信协议（如无线传输、蜂窝网络或Wi-Fi）对能源和延迟的影响，以及如何动态调整任务分发策略以适应实时变化的设备状态和网络条件。可能还包含了对不同计算任务类型（如实时任务、低延迟任务和能源效率优先任务）的处理方式，以及如何通过优化算法找到最优的能源消耗和延迟性能点。
此外，文中可能讨论了与其他传统计算模型（如云计算）相比，MEC的优势和局限性，以及未来的研究方向，比如考虑环境因素对能源和延迟的影响，或者开发更先进的自适应算法以进一步提升能源效率和用户体验。
这篇研究为移动网络运营商和设备制造商提供了一种实用的方法，帮助他们在一个日益复杂的环境中做出明智的决策，平衡能源消耗和用户需求的即时响应。这对于推动MEC技术的实际应用和发展具有重要的理论指导意义。
ZHANG et al.: ENERGY-LATENCY TRADEOFF FOR ENERGY-AWARE OFFLOADING IN MEC NETWORKS 2635
Fig. 1. Network model.
small cell eNodeBs (SeNBs) are overlaid by the MeNB. The
SeNBs are connected to the MeNB through wired link [27].
Each SeNB serves up to U
j
(j ∈{1, 2,...,M}) SMDs. In
order to reuse spectrum, we assume that multiple SeNBs
operate in the same frequency band, where the interference
between small cells exists. The bandwidth B is divided into
N channels. SMDs associate with the SeNBs in orthogonal
frequency-division multiple access (OFDMA), where the chan-
nel of each SMD in the same SeNB is orthogonal to others.
In this network, SMD i in SeNB j has a computation task
τ
i,j
={d
i,j
, c
i,j
, T
max
i,j
} to be completed, where d
i,j
is the size
of input data, c
i,j
denotes the total number of CPU cycles
required to accomplish the computation task, and T
max
i,j
denotes
the maximum tolerance latency. For each SMD, its task can
be executed either locally on itself or remotely in the MEC
server via computation offloading. Let s
i,j
be the offloading
decision of SMD i in cell j. If the SMD offloads its task to
the MEC server, s
i,j
= 1, otherwise, s
i,j
= 0.
1) Local Computing: We define f
l
i,j
as the computation abil-
ity (i.e., CPU cycles per second) of the SMD. When the task
τ
i,j
is executed locally on the SMD, the computation execution
time t
L
i,j
is
t
L
i,j
=
c
i,j
f
l
i,j
. (1)
The energy consumption of the SMD can be calculated as
e
L
i,j
= κ(f
l
i,j
)
2
c
i,j
(2)
where κ = 10
−26
, and it is a coefficient depending on the chip
architecture [28], [29].
Considering that f
l
i,j
affects the computation execution time
and energy consumptions simultaneously. It is allowed to
schedule the CPU-cycle frequency via dynamic voltage and
frequency scaling technology [22].
2) Edge Computing: When the input data is transmitted to
the MEC server through SeNB, the transmission expenditure
between the MEC sever and the SeNB is ignored [24], [30].
If the SMD accesses the SeNB on channel n, the achievable
uplink transmission rate can be expressed as
r
i,j,n
= wlog
2
1 +
p
i,j,n
h
i,j,n
σ
2
+ I
i,j,n
(3)
where w is the bandwidth, w = B/N. p
i,j,n
and h
i,j,n
are the
transmission power and the channel gain between the SMD
i and the SeNB j on channel n, respectively. σ
2
is the noise
power. I
i,j,n
denotes the interference of SMD i in cell j suffer-
ing from other SMDs in neighboring cells on the same channel
n, and it can be represented as
I
i,j,n
=
U
l
k=1
M
l=1,l=j
a
k,l,n
p
k,l,n
h
j
k,l,n
(4)
where l is the lth except the jth small cell, h
j
k,l,n
is the channel
gain from SMD k in cell l to cell j on channel n, and U
l
is
the number of SMDs in small cell l. Hence, the total uplink
transmission rate for SMD i in cell j can be obtained as
r
i,j
=
N
n=1
a
i,j,n
r
i,j,n
(5)
where a
i,j,n
∈{0, 1}. a
i,j,n
= 1 denotes the channel n is
assigned to SMD i in cell j to offload its task, otherwise,
a
i,j,n
= 0. Let f
C
denotes the CPU-cycle frequency of the
MEC server, which is fixed for the duration of computation
task execution [17], [22]. Then the total edge computing exe-
cution time of the task includes the transmission time and
computation time on MEC server, which can be given as
t
C
i,j
=
d
i,j
r
i,j
+
c
i,j
f
C
. (6)
The energy consumption of the SMD is expressed as
e
C
i,j
=
N
n=1
a
i,j,n
p
i,j,n
d
i,j
r
i,j
. (7)
Here, the time and energy consumption of outcome from
the MEC server to the SMD are ignored in this case, due to
the fact that the size of outcome data is much smaller than the
size of input data, which is similar to the studies [30], [31].
In the process of task execution, both execution latency and
energy consumption are vital for the SMDs, which attributes
to user experience and the battery energy limitation of SMDs.
In general, a weighting factor w
i,j
(w
i,j
∈ [0, 1]) will be
used to investigate the tradeoff between energy consumption
and latency, which can be defined by different SMDs for
the purpose of meeting the user-specific demands [8], [24].
Saving more energy or reducing latency can be implemented
by adjusting the weighting factor. However, we bring the resid-
ual energy rate r
E
i,j
of the battery into the weighting factor in
our model. It is defined as
w
i,j
= w
i,j
r
E
i,j
(8)
where r
E
i,j
= E
max
i,j
/E
total
. E
max
i,j
is the maximum residual energy
aware of the battery of SMD i in cell j and E
total
is the battery
capacity in Joules [26]. Unlike w
i,j
, r
E
i,j
is a definite value,
which reflects the real-time service condition of the battery.
According to (1) and (2), the overhead of the task locally
computed on SMD i in cell j, namely the weighted sum of
energy consumption and latency G
L
i,j
, can be defined as
G
L
i,j
= w
i,j
t
L
i,j
+ (1 − w
i,j
)αe
L
i,j
(9)
where α is the normalizing factor, which is introduced to
implement the unitless combination of energy consumption
剩余12页未读，继续阅读
2021-03-23 上传
2021-05-22 上传
2021-02-07 上传
2021-02-07 上传
2021-02-09 上传
2021-04-05 上传
2021-03-21 上传
2021-02-22 上传
2021-05-07 上传
weixin_38634323
- 粉丝: 7
- 资源: 898
我的内容管理
展开
最新资源
- Cucumber-JVM模板项目快速入门教程
- ECharts打造公司组织架构可视化展示
- DC Water Alerts 数据开放平台介绍
- 图形化编程打造智能家居控制系统
- 个人网站构建：使用CSS实现风格化布局
- 使用CANBUS控制LED灯柱颜色的Matlab代码实现
- ACTCMS管理系统安装与更新教程
- 快速查看IP地址及地理位置信息的View My IP插件
- Pandas库助力数据分析与编程效率提升
- Python实现k均值聚类音乐数据可视化分析
- formdotcom打造高效网络表单解决方案
- 仿京东套餐购买列表源码DYCPackage解析
- 开源管理工具orgParty：面向PartySur的多功能应用程序
- Flutter时间跟踪应用Time_tracker入门教程
- AngularJS实现自定义滑动项目及动作指南
- 掌握C++编译时打印：compile-time-printer的使用与原理
