延迟感知动态资源调配：实现混合云运营成本最小化
31 浏览量
更新于2024-08-27
收藏 945KB PDF 举报
"这篇研究论文探讨了如何在混合云环境中通过使用延迟感知的动态资源调配策略来最小化运营成本。作者Song Li、Yangfan Zhou、Lei Jiao、Xinya Yan、Xin Wang以及Michael Rung-Tsong Lyu提出了一种方法，旨在帮助软件即服务（SaaS）提供商更有效地处理用户请求的动态变化。
混合云计算模型被广泛推崇，因为它允许SaaS提供商将本地服务无缝扩展到公共云，从而利用本地服务器和公共云的计算能力共同处理用户工作负载。然而，尽管混合云可能比建立强大的私有云更具成本效益，但租赁费用和通信费用仍是混合云模式下的主要开支。
论文指出，当前技术尚未解决如何优化这些运营成本的问题。因此，该研究的目标是设计一种基于延迟感知的优化方案，以在满足服务质量（QoS）的同时，降低租用和通信成本。延迟感知的重要性在于，它考虑了资源调配对用户体验的影响，确保在降低成本的同时不牺牲用户体验。
论文可能会详细介绍以下几点：
1. **延迟感知模型**：构建一个能够量化延迟影响的数学模型，以评估不同资源分配决策对用户延迟的贡献。
2. **动态资源调配算法**：设计一种智能算法，能够在用户需求变化时实时调整本地和公共云资源的分配，以达到成本效率最大化。
3. **成本和性能权衡**：分析在各种业务场景下，如何在成本节省与性能保证之间找到最佳平衡点。
4. **实验评估**：通过模拟或真实环境的实验，验证所提出的策略在实际应用中的效果，对比现有解决方案，展示其优越性。
5. **未来研究方向**：讨论可能的改进点和未来研究的潜在领域，如更复杂的延迟模型、多云环境的适应性以及安全性考虑。
该论文为SaaS提供商提供了一个新的视角，即如何利用延迟感知技术在混合云环境中进行动态资源调配，以达到运营成本的最小化。这一研究对于推动混合云计算的实践应用具有重要意义，有助于企业更好地管理云资源，降低运维成本，提升服务质量。"
capacity via renting the VMs from a public IaaS cloud. This
is a typical hybrid cloud model.
The public IaaS cloud will in general provision three
types of VM services for the SaaS provider. The first is the
reserved VM service, which is long-term service with a fixed
VM numbers. We let T denote the minimum allowed rent-
ing period of such services. T is typically several days or
months [4], [28]. The second is the on-demand VM service,
where the number of VMs can be set instantaneously by the
SaaS provider. The last is the on-spot VM services, which the
SaaS provider can bid for. The price of the reserved service
per unit is typically the lowest. The on-demand one is often
the most expensive but it is charged in a pay-as-you-go
fashion. The price of the on-spot one is dynamic according
to the user bid (reflecting the user demand). The above is a
typical IaaS provision scheme for current public IaaS cloud
(e.g., the Amazon EC2).
Since renting the public cloud resources incurs monetary
cost, the SaaS provider should reduce such cost. The key
problem to the SaaS provider is how to decide the number
of VMs it need to rent so that the performance requirement
is satisfied and the cost is minimized. We use the symbols
R, D, and S to denote the numbers of VMs of the above
three services an SaaS provider will rent respectively. Also
let L denote the numbers of the VMs in the local servers the
SaaS provider owns. Let R
max
, D
max
, S
max
and L
max
denote
the upper bound of the numbers of each VM type, and let
W denote the lower bound of the numbers of all of VMs.
In the real-world scenario, the SaaS provide has to
make decisions on the number of the VMs in some
degrees of granularity. Si nce the allowed mi nimum rent-
ing period of reserved VMs is T ,weconsidertheSaaS
provider will d ivide its operation period into a sequence
of time intervals with length T and determine the number
of the reserved VMs at the beginning of each interval. We
name such a decision i nterval with length T acoarse-
grained decisio n interval.
The numbers of the other two types of VMs (the on-
demand ones and the on-spot ones) can be decided in a rela-
tively short-term manner. Without loss of generality, we
consider the operation time can be divided into many non-
overlapping small decision slots with length t. Let t
1
,
t
2
; ...; t
i
; ... (i is a non-negative integer) denote such slots
in sequence. The length of each t
i
is t. For example, in Ama-
zon EC2, t is one hour, which means that the SaaS provider
can make the decisions of the on-demand VMs and the on-
spot ones per hour. We name such a decision slots with
length t a fine-grained decision time slot.
Without loss of generality, suppose T ¼ mt. In other
words, the length of the coarse-grained decision interval is
m times of the fine-grained decision time slot. The SaaS pro-
vide can decide the number of the reserved VMs in the
beginning of time slot t
0
, t
m
, t
2m
; ...; t
jm
; ... where j is a
non-negative integer.
In this way, how to determine the numbers of the above
four types of VMs for the SaaS provider can be divided into
a two-scale decision process. Finally, Table I provides a
summary of the symbols used in this paper.
3.2 User Requests and SaaS Service Model
Consider the users access the SaaS provider with an arrival
rate ðt
i
Þ. Note that ðt
i
Þ can bear an arbitrary distribution
in realistic scenarios.
The user requests to the SaaS provider usually have
deadlines, allowing the request to wait f or a maximum
periods of time d
max
before it is scheduled. Consider the
requests can be stored in a queue denoted by Qðt
i
Þ in
each fine- grained decision time slot. In each find-grained
interval, the user requests in the queue are served with
the rat e mðt
i
Þ. Therefore, the queue has the following
update equation:
Qðt
i þ 1
Þ¼maxfQðt
i
Þmðt
i
Þ; 0gþðt
i
Þ: (1)
To serve the user requests, the SaaS provider allocates
each request a certain number of VMs according to the
request requirement, consisting of those in the local servers
and those in the public IaaS cloud. The service rate in each
time slot t
i
in a coarse-grained decision interval is
TABLE 1
Notation
Symbols Descriptions
T Minimum renting period of reserved VMs
t
i
The ith decision slot for the on-demand VMs and
the on-spot VMs
t The length of each decision slot for the on-demand
VMs and the on-spot VMs. T ¼ mt where m is an
integer.
Rðt
i
Þ The number of reserved VMs at time slot t
i
R
max
The upper bound of the number of the reserved
VMs
Dðt
i
Þ The number of on-demand VMs at time slot t
i
D
max
The upper bound of the number of the on-demand
VMs
Sðt
i
Þ The number of on-spot VMs at time slot t
i
S
max
The upper bound of the number of the on-spot
VMs
Lðt
i
Þ The number of VMs in the local servers at time
slot t
i
L
max
The upper bound of the number of the VMs in the
local servers
Wðt
i
Þ The lower bound of the number of VMs at time
slot t
i
ðt
i
Þ The arrival rate of user requests at time slot t
i
mðt
i
Þ The service rate of user request at time slot t
i
d
max
The maximum allowed queueing time of each
user request
Qðt
i
Þ The backlog of the user request queue
Q
max
The maximum of the queue backlog
Zðt
i
Þ The backlog of the virtual queue
Z
max
The maximum of the virtual queue backlog
f The cost of running the local servers
P
r
ðt
i
Þ The price of reserved VMs at t
i
P
d
ðt
i
Þ The price of on-demand VMs at t
i
P
s
ðt
i
Þ The price of on-spot VMs at t
i
Mðt
i
Þ The communication cost between VMs in the local
servers and those in the cloud
bðt
i
Þ The service response time at time slot t
i
b
max
The service response time upper bound requirment
The parameter bounding growing rate of the
virtual queue
V Parameter for the trade-off between queueing
delay and the cost
400 IEEE TRANSACTIONS ON SERVICES COMPUTING, VOL. 8, NO. 3, MAY/JUNE 2015
剩余11页未读，继续阅读
点击了解资源详情
点击了解资源详情
点击了解资源详情
2021-05-26 上传
2021-07-21 上传
2021-08-21 上传
2013-02-22 上传
点击了解资源详情
点击了解资源详情
weixin_38557896
- 粉丝: 0
- 资源: 971
我的内容管理
展开
最新资源
- JavaScript实现的高效pomodoro时钟教程
- CMake 3.25.3版本发布：程序员必备构建工具
- 直流无刷电机控制技术项目源码集合
- Ak Kamal电子安全客户端加载器-CRX插件介绍
- 揭露流氓软件：月息背后的秘密
- 京东自动抢购茅台脚本指南：如何设置eid与fp参数
- 动态格式化Matlab轴刻度标签 - ticklabelformat实用教程
- DSTUHack2021后端接口与Go语言实现解析
- CMake 3.25.2版本Linux软件包发布
- Node.js网络数据抓取技术深入解析
- QRSorteios-crx扩展：优化税务文件扫描流程
- 掌握JavaScript中的算法技巧
- Rails+React打造MF员工租房解决方案
- Utsanjan：自学成才的UI/UX设计师与技术博客作者
- CMake 3.25.2版本发布，支持Windows x86_64架构
- AR_RENTAL平台：HTML技术在增强现实领域的应用
