网络工程师故障诊断时最常犯的 10 个错误

排除网络和应用问题时有几个容易落入的陷阱。
This white paper describes 10 of these and how to avoid them.

  • 目录
  • Making assumptions
  • Rebooting
  • Upgrading
  • Validating
  • No baseline
  • Wireless challenges
  • Under-monitoring
  • Understanding core technologies
  • Laptop limits
  • 结论
 

100% uptime?Check.

10Gbps to the desktop?Getting there.

Problem free network and applications?Ha!In our dreams.

Until we reach network perfection (which, let's be honest, may never really happen) engineers will face problems with network systems and the applications they support.Whether the issue is slow performance, poor voice/video quality, dropped connections, or other events that plague networks today, engineers need to continually hone their troubleshooting skills to stay on top of these business efficiency-killers.Additionally, they need to avoid the common pitfalls that all too many network engineers fall into when troubleshooting problems.

Let’s look at a few examples.

1.假设问题的根本原因。

Let's face it:we humans make assumptions based on what we think we know.When a problem strikes, we can't help jumping to conclusions, especially if we have both time and experience in a particular network environment.However, making assumptions can be a huge mistake.They can lead to nonsensical network changes, costly upgrades, and baseless "improvements"- all with our fingers crossed, hoping the problem will go away.This troubleshooting mistake should be avoided at all costs.因此,在做这些下意识的决定前,我们应该先收集关于问题的资料。做出更改之前要完全了解问题中的主体、原因、地点、内容以及方式。Let fact guide every decision made.

 

2.“这种修复办法以前管用,我们再试试吧”故障诊断

Similar to mistake number one, this common response to network problems is also based on assumptions.We are all victims of our own experience, so it's easy to rely on our knowledge of what worked last time, thinking that the same will be true again.In many cases, a new problem will show the same symptoms as a previous one, but the root cause could be entirely different.

Before changing anything, make sure to isolate the problem domain to the network, server, application, or client.Clearly identify which component is to blame before trying the guess-and-change approach.Using tools that make use of SNMP, NetFlow, and packet capture, clearly isolate the problem to a layer before moving forward with the resolution.

3.通过重启解决问题

From home routers to 10G switches, almost all electronic devices need to be rebooted at one time or another.It's just a part of how things work today.However, in some IT environments, rebooting a device has become the standard for first-step troubleshooting.This is especially true if a device or server reboot has worked in the past.

If rebooting a device resolves a problem, the fix may only be temporary, requiring another reboot in the near future.Of course, a reboot could be required after a software upgrade, patch, or configuration change.However as a first-response to a network problem, repeatedly rebooting a device will only mask the real root cause.Prior to rebooting a device, collect as much information as possible.For example, is the access point still responding to current users?Is the server accepting new TCP connections?Is the switch CPU at 100% utilization?This information may steer engineers toward the real root cause, rather than the temporary fix.

 

4.通过升级解决问题

从 1Gbps 升级到 10Gbps 应该提高 10 倍的速度,对吗?

No.

Seldom is this the case.All too often, when faced with network problems - especially ones involving slow performance - network engineers are tempted to increase WAN bandwidth, upgrade switches or routers, or implement acceleration technologies.It’s no secret; none of these "fixes" are free.In fact, upgrading as a first-response to a problem can drain the budget, frustrate managers, reduce business productivity, and at worst, cost the job of a network engineer (yikes!).

Before implementing a new technology or upgrading a system/device/connection, there are several important questions to answer:Why are we convinced that this device/technology improvement will resolve the issue?What IS the original issue?Is the problem really rooted in network capacity or latency?

While it’s nice to have new gear on the network, it’s not pleasant to see the look on a manager’s face when an expensive solution fails to address a problem.Upgrading key systems is warranted from time to time, but be careful when upgrading a device as a troubleshooting step.


 

5.未验证便向用户交付新连接

We've all done it a million times.Unbox and configure a new switch, install it, patch in the uplink, connect the end user drop and watch the light go blinky-blinky.

Done, right?

No. There are several things that can impact the performance experienced by end users once they connect and get to work.Link negotiation, cable problems, interface hardware issues and other throughput killers can impact the connection.

Before officially delivering a link to an end user, it should be tested and validated.This includes measuring latency and throughput for each connection back to the core/data center.As we mentioned, most engineers will connect a link, look for a link light, send a ping and consider the link tested.However all of the issues described earlier would pass this test.Only a full performance test would validate the connection and reveal these problems before the users experience them.

 

6.未在网络正常运行时建立基准

When troubleshooting a problem, engineers often utilize monitoring tools to help them collect and interpret information about the network.Even though these tools can display an impressive amount of statistical data, it’s easy to get lost in the details if a “normal” baseline does not exist.

Before a problem strikes, effort should be made to properly baseline the network.This would include collecting traffic utilization and latency statistics on key network links, response time measurements on critical business applications, packet capture samples including typical conversations and protocols, and a complete wireless assessment.These reports will assist network engineers when an issue arises since they will know what "normal" is.


 

7.缺少无线工具和经验

Wireless can be a real pain, especially as more end user devices ditch the cable and go 100% Wi-Fi.This trend, as well as the increase in the voice and video applications these devices demand, has greatly elevated the scope and complexity of wireless environments.Even when these systems are implemented and maintained by seasoned RF experts, clients can still experience poor performance, network disconnections, and other frustrating issues.

因为无线环境很容易受到性能问题的干扰,所以出现新问题时通常会首先归咎于它。Many network engineers point the finger at the Wi-Fi simply because it is an area of the network they don't fully understand or lack the tools to analyze.Rather than have a huge network blind spot, network managers should invest in both tools and training to get engineers up to speed on wireless, equipping them to respond to problems in this domain.

 

8.未对网络提供足够监控

Problems that engineers face today are complex, intermittent, and manage to hide in the shadows of the system.在过去,监控网络只需要使用一个上/下 ping 工具。这已经发生了很大的变化。

解决现在的问题需要监控同时感知网络和应用的系统,利用 SNMP、NetFlow 以及数据包捕捉了解一切信息。这些系统需要监控应用 24/7/365 保证实时捕捉间歇性问题,而不能发生监控系统对事件视而不见的情况。

 

9.误解核心技术的运作方式

What do spanning-tree, ARP, auto-negotiation, ICMP redirects and IP fragmentation have in common?

They are all old (20+ years for each) and absolutely critical for network operation.Well, maybe not IP fragmentation in every case but it was worth the mention.Network engineers need to ensure that they understand the core technologies that their state-of-the-art systems are built on.When prepping for that next vendor certification exam, don't leave out the protocols and technologies that still have a hand in keeping things running this year, and beyond.


 

10.使用笔记本硬件捕获数据包

调查问题时,数据包捕捉和跟踪文件解释是深入分析详情的黄金标准。与仅仅责备网络并把问题抛给他人相比,这种分析方法才是查找问题的根本原因的关键。

When it comes to packet collection, a common mistake made by network engineers today is misunderstanding the limits of the hardware they are using to capture.以 Wireshark 为例,这一开源工具被世界各地的工程师了解和喜爱,是下载最多的网络工具。但是,大部分人在笔记本或未测试的硬件上使用此工具时,无法跟上高速流量。实际上,速度超过 100Mbps 后,大部分标准笔记本就难以流畅地进行捕捉了!

在数据中心环境中收集数据包之前先了解收集时使用的硬件的限制。Missing packets from trace files can easily lead an engineer astray, increasing the time to resolution of a nagging problem.

 

结论

This is not an exhaustive list - there are other pitfalls that engineers of all experience levels fall into from time to time.With a little bit of preparation and awareness of some common mistakes, engineers can reduce time to resolution, trim frustration, reduce costs or unnecessary expense, and avoid the headaches brought on when troubleshooting network problems.

 
 
Powered By OneLink