优化SDN交换机：缓冲区策略的效益与机制设计
102 浏览量
更新于2024-08-27
收藏 1.19MB PDF 举报
本文探讨了在软件定义网络（Software-Defined Networking, SDN）交换机中应用缓冲区的策略，其核心关注点在于提升网络动态管理效率，特别是对于突发故障处理和流量精细化调度。SDN的核心优势在于通过集中式的控制平面与分布式的数据平面分离，使得网络配置和管理变得更加灵活。然而，这种分离架构带来的一个挑战是控制与数据之间的通信开销，尤其是由于数据包转发规则的请求和响应过程中频繁的交互。
传统上，当SDN交换机接收到未知或不符合预先定义规则的数据包（miss-match packets）时，会发送请求至控制器寻求转发指示，这会导致显著的通信延迟和带宽占用。为了减少这一问题，文章提出了将缓冲区机制引入SDN交换机的设计。通过在交换机上存储并暂时处理这些未知数据包，缓冲区可以缓存这些请求，直到控制器做出响应后再进行转发，从而减少了不必要的通信次数。
本研究首先对使用缓冲区的潜在益处进行了深入分析。这包括但不限于：
1. 提高网络吞吐量：通过减少不必要的控制请求，缓冲区可以降低数据包处理的实时响应时间，释放控制器资源，提高整体网络性能。
2. 减少网络延迟：通过本地处理部分请求，延迟问题得以缓解，提高了用户体验和网络稳定性。
3. 节省带宽：通过在交换机层面缓存和处理数据，减少了与控制器之间的通信流量，减轻了网络拥塞。
4. 提高可靠性：缓冲机制有助于在控制器短暂故障或响应延迟时维持网络连通性，提高了网络的可用性。
接着，文章详细探讨了如何设计和实现这样的缓冲机制。这可能涉及以下几个关键方面：
1. 缓冲区大小优化：根据网络流量预测和控制器处理能力，合理设置缓冲区大小，以平衡存储效率和处理能力。
2. 缓存策略：选择合适的缓存策略，如LRU（Least Recently Used，最近最少使用）或LFU（Least Frequently Used，最不经常使用），以决定何时丢弃缓存中的数据包。
3. 优先级机制：区分不同类型的流量，可能需要为优先级高的流量提供更大的缓存空间，确保关键业务不受影响。
4. 高效更新策略：当控制器更新转发规则时，如何快速且高效地更新缓冲区中的信息，以最小化切换期间的数据丢失。
5. 监控和调整：定期评估缓冲机制的效果，根据实际网络状况动态调整参数，以保持最佳性能。
总结来说，本文为SDN交换机的通信开销问题提供了一个创新的解决方案，通过引入缓冲机制来提升网络效率和可靠性。通过深入的收益分析和机制设计，研究人员展示了将缓冲技术整合到SDN架构中的可行性和潜在价值，为未来网络设计和优化提供了有价值的参考。
2168-7161 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCC.2018.2846620, IEEE
Transactions on Cloud Computing
3
depends on applied actions and firmware. Huang et al. [12]
investigated three commercial OpenFlow switches and pointed
out that control path delays and flow table designs affect
switching performance. Tai et al. [14] uncovered the source
of forwarding latency caused by forwarding rule insertions on
data plane. Xu et al. [31] proposed two cost-optimized flow
statistics collection schemes using wildcard-based requests,
which could reduce the bandwidth overhead and switch pro-
cessing delay. He et al. [8, 9] studied four SDN switches
and showed the latencies underlying the generation of request
messages and execution of forwarding rules. Different from
previous works, we try to analyze the benefits of adopting
switch buffer through an in-depth measurement study, which
could make a complementary exploitation of SDN switches.
Observed from the analysis results, we find that switch buffer
can reduce the communication overhead between the switches
and the controller.
The root cause of communication overhead stems from
the size limitation of SDN flow tables. Rules in flow tables
have to be updated adapting to network dynamics, i.e., rules
for inactive flows will be kicked out and replaced by rules
for active flows. As a result, packets cannot always match
the rules of the flow tables. Therefore, optimizing flow ta-
ble utilization (such as caching more rules, updating rules
quickly, etc.) can reduce the requests sent to the controller
indirectly. Luo et al. [16] shrank the flow table size and
provided practical methods to achieve fast flow table updates.
Li et al. [17] proposed an efficient flow-driven rule caching
algorithm to optimize the SDN switch cache replacement.
CacheFlow [18] spliced long dependency chains to cache
smaller groups of rules while preserving the per-rule traffic
counts. FlowShadow [19] achieved fast packet processing and
supports uninterrupted update by caching microflows. Zhang
et al. [32] propose a delay-guaranteed approach called D
3
G
to reduce the latency of chained services while obtain fairness
across all the workloads, by means of designing a latency
estimation algorithm and a feedback scheme. Yan et al. [29]
proposed a new rule caching scheme, as well as an adaptive
cache management method. The mechanism can reduce the
cache miss rate by one order of magnitude and the control
path bandwidth usage by a half. Different from these studies,
this paper exploits the benefits of switch buffer to directly
reduce the communication overhead. Buffer has been adopted
to improve performance and achieve QoS guarantee in legacy
switches [20, 21]. However, it receives little notice in SDN.
III. PROBLEM AND EXPERIMENT DESCRIPTION
In this section, we first describe the problem and then show
the measurement methodology.
A. Problem Description
A flow contains many packets of {p
1
, p
2
, ..., p
n
} arriving
at {t
1
, t
2
, ..., t
n
}. If p
1
matches a rule of the flow table, it
will be forwarded at a line rate. Otherwise, the switch will
generate a pkt in message sent to the controller. After the
controller decides how to forward the packet, it will send a pair
of control operation messages (flow mod and pkt out) to the
TABLE I: Configurations of Experimental Devices
Device Name CPU Cores RAM NIC
Host
1
3.3GHZ 4 4GB 1×100Mbps
Host
2
3.3GHZ 4 4GB 1×100Mbps
Open vSwitch 3.3GHZ 4 4GB 3×100Mbps
Floodlight 3.3GHZ 4 4GB 1×100Mbps
Fig. 1: Topography of the Experimental Platform.
switch: flow mod message carries the forwarding rule that will
be installed in the switch; pkt out message instructs to directly
forward the miss-match packet through a specified interface of
the switch. The time of the flow mod message taking effect is
t
e
. If t
e
> t
2
, p
2
will trigger another pkt in message. In the
worst case, if t
e
> t
n
, n pkt in messages will be triggered.
When massive packets fails to match any rules of the flow
tables simultaneously, a great deal of request messages will
be sent to the controller. Moreover, corresponding control
operation messages will be sent back to the switch. Such
communication overhead needs to be reduced by the following
three reasons.
1) Control path may share the same physical links with the
data path. When the physical links carry heavy data traffic,
control messages may be congested. 2) Even if we reserve the
bandwidth or increase the priority for control traffic, we still
have the requirements of reducing the control messages to
relieve the load on the centralized controller. 3) Concurrent
switch activities, i.e., generating control request messages
and handling control operation messages will increase the
communication delay between the switch and the controller
[9]. This is caused by the competition of the limited resources
of the switch. So it is important to keep the control traffic at
a low level.
B. Experiment Description
Fig. 1 shows our experiments setup. Open vSwitch(OVS)
[22] is an open source OpenFlow virtual switch. Floodlight
[23] is an open source SDN controller. We run OVS and
Floodlight on two commodity PCs respectively. Table I shows
the configurations of the experimental devices. Host
1
and
Host
2
connect to OVS with 100Mbps interfaces. We run
pktgen [24] on Host
1
to generate traffic at rates of 5Mbps
- 100Mbps with the Ethernet frame size of 1000 Bytes. We
run tcpdump [25] to listen on the interfaces that are connected
to the hosts and the controller respectively.
In this paper, we utilize switch buffer to reduce the com-
munication overhead from the following two ways: (1) reduce
Authorized licensed use limited to: Northeastern University. Downloaded on March 24,2020 at 03:43:30 UTC from IEEE Xplore. Restrictions apply.
剩余11页未读，继续阅读
2022-05-28 上传
2022-05-31 上传
2021-02-05 上传
2021-02-05 上传
2021-02-05 上传
点击了解资源详情
点击了解资源详情
点击了解资源详情
weixin_38665411
- 粉丝: 8
- 资源: 936
我的内容管理
展开
最新资源
- WordPress作为新闻管理面板的实现指南
- NPC_Generator：使用Ruby打造的游戏角色生成器
- MATLAB实现变邻域搜索算法源码解析
- 探索C++并行编程：使用INTEL TBB的项目实践
- 玫枫跟打器：网页版五笔打字工具，提升macOS打字效率
- 萨尔塔·阿萨尔·希塔斯：SATINDER项目解析
- 掌握变邻域搜索算法：MATLAB代码实践
- saaraansh: 简化法律文档，打破语言障碍的智能应用
- 探索牛角交友盲盒系统：PHP开源交友平台的新选择
- 探索Nullfactory-SSRSExtensions: 强化SQL Server报告服务
- Lotide：一套JavaScript实用工具库的深度解析
- 利用Aurelia 2脚手架搭建新项目的快速指南
- 变邻域搜索算法Matlab实现教程
- 实战指南：构建高效ES+Redis+MySQL架构解决方案
- GitHub Pages入门模板快速启动指南
- NeonClock遗产版：包名更迭与应用更新
