mirror of
https://github.com/LCTT/TranslateProject.git
synced 2025-02-25 00:50:15 +08:00
TSL
This commit is contained in:
Xingyu Wang
parent
01a44359a5
commit
556b589822
@ -1,267 +0,0 @@
|
||||
[#]: subject: (Identify Linux performance bottlenecks using open source tools)
|
||||
[#]: via: (https://opensource.com/article/21/3/linux-performance-bottlenecks)
|
||||
[#]: author: (Howard Fosdick https://opensource.com/users/howtech)
|
||||
[#]: collector: (lujun9972)
|
||||
[#]: translator: (wxy)
|
||||
[#]: reviewer: ( )
|
||||
[#]: publisher: ( )
|
||||
[#]: url: ( )
|
||||
|
||||
Identify Linux performance bottlenecks using open source tools
|
||||
======
|
||||
Not long ago, identifying hardware bottlenecks required deep expertise.Today's open source GUI performance monitors make it pretty simple.
|
||||
![Lightning in a bottle][1]
|
||||
|
||||
Computers are integrated systems that only perform as fast as their slowest hardware component. If one component is less capable than the others—if it falls behind and can't keep up—it can hold your entire system back. That's a _performance bottleneck_. Removing a serious bottleneck can make your system fly.
|
||||
|
||||
This article explains how to identify hardware bottlenecks in Linux systems. The techniques apply to both personal computers and servers. My emphasis is on PCs—I won't cover server-specific bottlenecks in areas such as LAN management or database systems. Those often involve specialized tools.
|
||||
|
||||
I also won't talk much about solutions. That's too big a topic for this article. Instead, I'll write a follow-up article with performance tweaks.
|
||||
|
||||
I'll use only open source graphical user interface (GUI) tools to get the job done. Most articles on Linux bottlenecking are pretty complicated. They use specialized commands and delve deep into arcane details.
|
||||
|
||||
The GUI tools that open source offers make identifying many bottlenecks simple. My goal is to give you a quick, easy approach that you can use anywhere.
|
||||
|
||||
### Where to start
|
||||
|
||||
A computer consists of six key hardware resources:
|
||||
|
||||
* Processors
|
||||
* Memory
|
||||
* Storage
|
||||
* USB ports
|
||||
* Internet connection
|
||||
* Graphics processor
|
||||
|
||||
|
||||
|
||||
Should any one resource perform poorly, it can create a performance bottleneck. To identify a bottleneck, you must monitor these six resources.
|
||||
|
||||
Open source offers a plethora of tools to do the job. I'll use the [GNOME System Monitor][2]. Its output is easy to understand, and you can find it in most repositories.
|
||||
|
||||
Start it up and click on the **Resources** tab. You can identify many performance problems right off.
|
||||
|
||||
![System Monitor - Resources Panel ][3]
|
||||
|
||||
Fig. 1. System Monitor spots problems. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
The **Resources** panel displays three sections: **CPU History**, **Memory and Swap History**, and **Network History**. A quick glance tells you immediately whether your processors are swamped, or your computer is out of memory, or you're using up all your internet bandwidth.
|
||||
|
||||
I'll explore these problems below. For now, check the System Monitor first when your computer slows down. It instantly clues you in on the most common performance problems.
|
||||
|
||||
Now let's explore how to identify bottlenecks in specific areas.
|
||||
|
||||
### How to identify processor bottlenecks
|
||||
|
||||
To spot a bottleneck, you must first know what hardware you have. Open source offers several tools for this purpose. I like [HardInfo][5] because its screens are easy to read and it's widely popular.
|
||||
|
||||
Start up HardInfo. Its **Computer -> Summary** panel identifies your CPU and tells you about its cores, threads, and speeds. It also identifies your motherboard and other computer components.
|
||||
|
||||
![HardInfo Summary Panel][6]
|
||||
|
||||
Fig. 2. HardInfo shows hardware details. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
HardInfo reveals that this computer has one physical CPU chip. That chip contains two processors, or cores. Each core supports two threads, or logical processors. That's a total of four logical processors—exactly what System Monitor's CPU History section showed in Fig. 1.
|
||||
|
||||
A _processor bottleneck_ occurs when processors can't respond to requests for their time. They're already busy.
|
||||
|
||||
You can identify this when System Monitor shows logical processor utilization at over 80% or 90% for a sustained period. Here's an example where three of the four logical processors are swamped at 100% utilization. That's a bottleneck because it doesn't leave much CPU for any other work.
|
||||
|
||||
![System Monitor processor bottleneck][7]
|
||||
|
||||
Fig. 3. A processor bottleneck. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
#### Which app is causing the problem?
|
||||
|
||||
You need to find out which program(s) is consuming all that CPU. Click on System Monitor's **Processes** tab. Then click on the **% CPU** header to sort the processes by how much CPU they're consuming. You'll see which apps are throttling your system.
|
||||
|
||||
![System Monitor Processes panel][8]
|
||||
|
||||
Fig. 4. Identifying the offending processes. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
The top three processes each consume 24% of the _total_ CPU resource. Since there are four logical processors, this means each consumes an entire processor. That's just as Fig. 3 shows.
|
||||
|
||||
The **Processes** panel identifies a program named **analytical_AI** as the culprit. You can right-click on it in the panel to see more details on its resource consumption, including memory use, the files it has open, its input/output details, and more.
|
||||
|
||||
If your login has administrator privileges, you can manage the process. You can change its priority and stop, continue, end, or kill it. So, you could immediately resolve your bottleneck here.
|
||||
|
||||
![System Monitor managing a process][9]
|
||||
|
||||
Fig. 5. Right-click on a process to manage it. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
How do you fix processing bottlenecks? Beyond managing the offending process in real time, you could prevent the bottleneck from happening. For example, you might substitute another app for the offender, work around it, change your behavior when using that app, schedule the app for off-hours, address an underlying memory issue, performance-tweak the app or your system software, or upgrade your hardware. That's too much to cover here, so I'll explore those options in my next article.
|
||||
|
||||
#### Common processor bottlenecks
|
||||
|
||||
You'll encounter several common bottlenecks when monitoring your CPUs with System Monitor.
|
||||
|
||||
Sometimes one logical processor is bottlenecked while all the others are at low utilization. This means you have an app that's not coded smartly enough to take advantage of more than one logical processor, and it's maxed out the one it's using. That app will take longer to finish than it would if it used more processors. On the other hand, at least it leaves your other processors free for other work and doesn't take over your computer.
|
||||
|
||||
You might also see a logical processor stuck forever at 100% utilization. Either it's very busy, or a process is hung. The way to tell if it's hung is if the process never does any disk activity (as the System Monitor **Processes** panel will show).
|
||||
|
||||
Finally, you might notice that when all your processors are bottlenecked, your memory is fully utilized, too. Out-of-memory conditions sometimes cause processor bottlenecks. In this case, you want to solve the underlying memory problem, not the symptomatic CPU issue.
|
||||
|
||||
### How to identify memory bottlenecks
|
||||
|
||||
Given the large amount of memory in modern PCs, memory bottlenecks are much less common than they once were. Yet you can still run into them if you run memory-intensive programs, especially if you have a computer that doesn't contain much random access memory (RAM).
|
||||
|
||||
Linux [uses memory][10] both for programs and to cache disk data. The latter speeds up disk data access. Linux can reclaim that memory any time it needs it for program use.
|
||||
|
||||
The System Monitor's **Resources** panel displays your total memory and how much of it is used. In the **Processes** panel, you can see individual processes' memory use.
|
||||
|
||||
Here's the portion of the System Monitor **Resources** panel that tracks aggregate memory use:
|
||||
|
||||
![System Monitor memory bottleneck][11]
|
||||
|
||||
Fig. 6. A memory bottleneck. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
To the right of Memory, you'll notice [Swap][12]. This is disk space Linux uses when it runs low on memory. It writes memory to disk to continue operations, effectively using swap as a slower extension to your RAM.
|
||||
|
||||
The two memory performance problems you'll want to look out for are:
|
||||
|
||||
> 1. Memory appears largely used, and you see frequent or increasing activity on the swap space.
|
||||
> 2. Both memory and swap are largely used up.
|
||||
>
|
||||
|
||||
|
||||
Situation 1 means slower performance because swap is always slower than memory. Whether you consider it a performance problem depends on many factors (e.g., how active your swap space is, its speed, your expectations, etc.). My opinion is that anything more than token swap use is unacceptable for a modern personal computer.
|
||||
|
||||
Situation 2 is where both memory and swap are largely in use. This is a _memory bottleneck._ The computer becomes unresponsive. It could even fall into a state of _thrashing_, where it accomplishes little more than memory management.
|
||||
|
||||
Fig. 6 above shows an old computer with only 2GB of RAM. As memory use surpassed 80%, the system started writing to swap. Responsiveness declined. This screenshot shows over 90% memory use, and the computer is unusable.
|
||||
|
||||
The ultimate answer to memory problems is to either use less of it or buy more. I'll discuss solutions in my follow-up article.
|
||||
|
||||
### How to identify storage bottlenecks
|
||||
|
||||
Storage today comes in several varieties of solid-state and mechanical hard disks. Device interfaces include PCIe, SATA, Thunderbolt, and USB. Regardless of which type of storage you have, you use the same procedure to identify disk bottlenecks.
|
||||
|
||||
Start with System Monitor. Its **Processes** panel displays the input/output rates for individual processes. So you can quickly identify which processes are doing the most disk I/O.
|
||||
|
||||
But the tool doesn't show the _aggregate data transfer rate per disk._ You need to see the total load on a specific disk to determine if that disk is a storage bottleneck.
|
||||
|
||||
To do so, use the [atop][13] command. It's available in most Linux repositories.
|
||||
|
||||
Just type `atop` at the command-line prompt. The output below shows that device `sdb` is `busy 101%`. Clearly, it's reached its performance limit and is restricting how fast your system can get work done.
|
||||
|
||||
![atop disk bottleneck][14]
|
||||
|
||||
Fig. 7. The atop command identifies a disk bottleneck. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
Notice that one of the CPUs is waiting on the disk to do its job 85% of the time (`cpu001 w 85%`). This is typical when a storage device becomes a bottleneck. In fact, many look first at CPU I/O waits to spot storage bottlenecks.
|
||||
|
||||
So, to easily identify a storage bottleneck, use the `atop` command. Then use the **Processes** panel on System Monitor to identify the individual processes that are causing the bottleneck.
|
||||
|
||||
### How to identify USB port bottlenecks
|
||||
|
||||
Some people use their USB ports all day long. Yet, they never check if those ports are being used optimally. Whether you plug in an external disk, a memory stick, or something else, you'll want to verify that you're getting maximum performance from your USB-connected devices.
|
||||
|
||||
This chart shows why. Potential USB data transfer rates vary _enormously_.
|
||||
|
||||
![USB standards][15]
|
||||
|
||||
Fig. 8. USB speeds vary a lot. (Howard Fosdick, based on figures provided by [Tripplite][16] and [Wikipedia][17], [CC BY-SA 4.0][4])
|
||||
|
||||
HardInfo's **USB Devices** tab displays the USB standards your computer supports. Most computers offer more than one speed. How can you tell the speed of a specific port? Vendors color-code them, as shown in the chart. Or you can look in your computer's documentation.
|
||||
|
||||
To see the actual speeds you're getting, test by using the open source [GNOME Disks][18] program. Just start up GNOME Disks, select its **Benchmark Disk** feature, and run a benchmark. That tells you the maximum real speed you'll get for a port with the specific device plugged into it.
|
||||
|
||||
You may get different transfer speeds for a port, depending on which device you plug into it. Data rates depend on the particular combination of port and device.
|
||||
|
||||
For example, a device that could fly at 3.1 speed will use a 2.0 port—at 2.0 speed—if that's what you plug it into. (And it won't tell you it's operating at the slower speed!) Conversely, if you plug a USB 2.0 device into a 3.1 port, it will work, but at the 2.0 speed. So to get fast USB, you must ensure both the port and the device support it. GNOME Disks gives you the means to verify this.
|
||||
|
||||
To identify a USB processing bottleneck, use the same procedure you did for solid-state and hard disks. Run the `atop` command to spot a USB storage bottleneck. Then, use System Monitor to get the details on the offending process(es).
|
||||
|
||||
### How to identify internet bandwidth bottlenecks
|
||||
|
||||
The System Monitor **Resources** panel tells you in real time what internet connection speed you're experiencing (see Fig. 1).
|
||||
|
||||
There are [great Python tools out there][19] to test your maximum internet speed, but you can also test it on websites like [Speedtest][20], [Fast.com][21], and [Speakeasy][22]. For best results, close everything and run _only_ the speed test; turn off your VPN; run tests at different times of day; and compare the results from several testing sites.
|
||||
|
||||
Then compare your results to the download and upload speeds that your vendor claims you're getting. That way, you can confirm you're getting the speeds you're paying for.
|
||||
|
||||
If you have a separate router, test with and without it. That can tell you if your router is a bottleneck. If you use WiFi, test with it and without it (by directly cabling your laptop to the modem). I've often seen people complain about their internet vendor when what they actually have is a WiFi bottleneck they could fix themselves.
|
||||
|
||||
If some program is consuming your entire internet connection, you want to know which one. Find it by using the `nethogs` command. It's available in most repositories.
|
||||
|
||||
The other day, my System Monitor suddenly showed my internet access spiking. I just typed `nethogs` in the command line, and it instantly identified the bandwidth consumer as a Clamav antivirus update.
|
||||
|
||||
![Nethogs][23]
|
||||
|
||||
Fig. 9. Nethogs identifies bandwidth consumers. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
### How to identify graphics processing bottlenecks
|
||||
|
||||
If you plug your monitor into the motherboard in the back of your desktop computer, you're using _onboard graphics_. If you plug it into a card in the back, you have a dedicated graphics subsystem. Most call it a _video card_ or _graphics card._ For desktop computers, add-in cards are typically more powerful and more expensive than motherboard graphics. Laptops always use onboard graphics.
|
||||
|
||||
HardInfo's **PCI Devices** panel tells you about your graphics processing unit (GPU). It also displays the amount of dedicated video memory you have (look for the memory marked "prefetchable").
|
||||
|
||||
![Video Chipset Information][24]
|
||||
|
||||
Fig. 10. HardInfo provides graphics processing information. (Howard Fosdick, [CC BY-SA 4.0][4])
|
||||
|
||||
CPUs and GPUs work [very closely][25] together. To simplify, the CPU prepares frames for the GPU to render, then the GPU renders the frames.
|
||||
|
||||
A _GPU bottleneck_ occurs when your CPUs are waiting on a GPU that is 100% busy.
|
||||
|
||||
To identify this, you need to monitor CPU and GPU utilization rates. Open source monitors like [Conky][26] and [Glances][27] do this if their extensions work with your graphics chipset.
|
||||
|
||||
Take a look at this example from Conky. You can see that this system has a lot of available CPU. The GPU is only 25% busy. Imagine if that GPU number were instead near 100%. Then you'd know that the CPUs were waiting on the GPU, and you'd have a GPU bottleneck.
|
||||
|
||||
![Conky CPU and GPU monitoring][28]
|
||||
|
||||
Fig. 11. Conky displays CPU and GPU utilization. (Image courtesy of [AskUbuntu forum][29])
|
||||
|
||||
On some systems, you'll need a vendor-specific tool to monitor your GPU. They're all downloadable from GitHub and are described in this article on [GPU monitoring and diagnostic command-line tools][30].
|
||||
|
||||
### Summary
|
||||
|
||||
Computers consist of a collection of integrated hardware resources. Should any of them fall way behind the others in its workload, it creates a performance bottleneck. That can hold back your entire system. You need to be able to identify and correct bottlenecks to achieve optimal performance.
|
||||
|
||||
Not so long ago, identifying bottlenecks required deep expertise. Today's open source GUI performance monitors make it pretty simple.
|
||||
|
||||
In my next article, I'll discuss specific ways to improve your Linux PC's performance. Meanwhile, please share your own experiences in the comments.
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
|
||||
via: https://opensource.com/article/21/3/linux-performance-bottlenecks
|
||||
|
||||
作者：[Howard Fosdick][a]
|
||||
选题：[lujun9972][b]
|
||||
译者：[译者ID](https://github.com/译者ID)
|
||||
校对：[校对者ID](https://github.com/校对者ID)
|
||||
|
||||
本文由 [LCTT](https://github.com/LCTT/TranslateProject) 原创编译，[Linux中国](https://linux.cn/) 荣誉推出
|
||||
|
||||
[a]: https://opensource.com/users/howtech
|
||||
[b]: https://github.com/lujun9972
|
||||
[1]: https://opensource.com/sites/default/files/styles/image-full-size/public/lead-images/BUSINESS_lightning.png?itok=wRzjWIlm (Lightning in a bottle)
|
||||
[2]: https://wiki.gnome.org/Apps/SystemMonitor
|
||||
[3]: https://opensource.com/sites/default/files/uploads/1_system_monitor_resources_panel.jpg (System Monitor - Resources Panel )
|
||||
[4]: https://creativecommons.org/licenses/by-sa/4.0/
|
||||
[5]: https://itsfoss.com/hardinfo/
|
||||
[6]: https://opensource.com/sites/default/files/uploads/2_hardinfo_summary_panel.jpg (HardInfo Summary Panel)
|
||||
[7]: https://opensource.com/sites/default/files/uploads/3_system_monitor_100_processor_utilization.jpg (System Monitor processor bottleneck)
|
||||
[8]: https://opensource.com/sites/default/files/uploads/4_system_monitor_processes_panel.jpg (System Monitor Processes panel)
|
||||
[9]: https://opensource.com/sites/default/files/uploads/5_system_monitor_manage_a_process.jpg (System Monitor managing a process)
|
||||
[10]: https://www.networkworld.com/article/3394603/when-to-be-concerned-about-memory-levels-on-linux.html
|
||||
[11]: https://opensource.com/sites/default/files/uploads/6_system_monitor_out_of_memory.jpg (System Monitor memory bottleneck)
|
||||
[12]: https://opensource.com/article/18/9/swap-space-linux-systems
|
||||
[13]: https://opensource.com/life/16/2/open-source-tools-system-monitoring
|
||||
[14]: https://opensource.com/sites/default/files/uploads/7_atop_storage_bottleneck.jpg (atop disk bottleneck)
|
||||
[15]: https://opensource.com/sites/default/files/uploads/8_usb_standards_speeds.jpg (USB standards)
|
||||
[16]: https://www.samsung.com/us/computing/memory-storage/solid-state-drives/
|
||||
[17]: https://en.wikipedia.org/wiki/USB
|
||||
[18]: https://wiki.gnome.org/Apps/Disks
|
||||
[19]: https://opensource.com/article/20/1/internet-speed-tests
|
||||
[20]: https://www.speedtest.net/
|
||||
[21]: https://fast.com/
|
||||
[22]: https://www.speakeasy.net/speedtest/
|
||||
[23]: https://opensource.com/sites/default/files/uploads/9_nethogs_bandwidth_consumers.jpg (Nethogs)
|
||||
[24]: https://opensource.com/sites/default/files/uploads/10_hardinfo_video_card_information.jpg (Video Chipset Information)
|
||||
[25]: https://www.wepc.com/tips/cpu-gpu-bottleneck/
|
||||
[26]: https://itsfoss.com/conky-gui-ubuntu-1304/
|
||||
[27]: https://opensource.com/article/19/11/monitoring-linux-glances
|
||||
[28]: https://opensource.com/sites/default/files/uploads/11_conky_cpu_and_gup_monitoring.jpg (Conky CPU and GPU monitoring)
|
||||
[29]: https://askubuntu.com/questions/387594/how-to-measure-gpu-usage
|
||||
[30]: https://www.cyberciti.biz/open-source/command-line-hacks/linux-gpu-monitoring-and-diagnostic-commands/
|
||||
@ -0,0 +1,265 @@
|
||||
[#]: subject: (Identify Linux performance bottlenecks using open source tools)
|
||||
[#]: via: (https://opensource.com/article/21/3/linux-performance-bottlenecks)
|
||||
[#]: author: (Howard Fosdick https://opensource.com/users/howtech)
|
||||
[#]: collector: (lujun9972)
|
||||
[#]: translator: (wxy)
|
||||
[#]: reviewer: ( )
|
||||
[#]: publisher: ( )
|
||||
[#]: url: ( )
|
||||
|
||||
使用开源工具识别 Linux 的性能瓶颈
|
||||
======
|
||||
|
||||
> 不久前，识别硬件瓶颈还需要深厚的专业知识。今天的开源 GUI 性能监视器使它变得相当简单。
|
||||
|
||||
![瓶子里的闪电][1]
|
||||
|
||||
计算机是一个集成的系统，它的性能取决于最慢的硬件组件。如果一个组件的能力比其他组件差，落后不能跟上，它就会拖累你的整个系统。这就是一个 _性能瓶颈_。消除一个严重的瓶颈可以使你的系统飞起来。
|
||||
|
||||
本文解释了如何识别 Linux 系统中的硬件瓶颈。这些技术同时适用于个人电脑和服务器。我强调的是个人电脑 —— 我不会涉及局域网管理或数据库系统等领域的服务器特定瓶颈。这些通常涉及专门的工具。
|
||||
|
||||
我也不会多谈解决方案。这对本文来说是个太大的话题。相反，我将写一篇关于性能调整的后续文章。
|
||||
|
||||
我将只使用开源的图形用户界面（GUI）工具来完成这项工作。大多数关于 Linux 瓶颈的文章都相当复杂。它们使用专门的命令，并深入研究神秘的细节。
|
||||
|
||||
开源提供的 GUI 工具使得识别许多瓶颈变得简单。我的目标是给你一个快速、简单的方法，你可以在任何地方使用。
|
||||
|
||||
### 从哪里开始
|
||||
|
||||
一台计算机由六个关键的硬件资源组成。
|
||||
|
||||
* 处理器
|
||||
* 内存
|
||||
* 存储器
|
||||
* USB 端口
|
||||
* 互联网连接
|
||||
* 图形处理器
|
||||
|
||||
如果任何一个资源表现不佳，就会产生一个性能瓶颈。为了识别瓶颈，你必须监测这六种资源。
|
||||
|
||||
开源提供了大量的工具来完成这项工作。我将使用 [GNOME 系统监视器][2]。它的输出很容易理解，而且你可以在大多数软件库中找到它。
|
||||
|
||||
启动它并点击“资源”标签。你可以马上发现许多性能问题。
|
||||
|
||||
![系统监控-资源面板][3]
|
||||
|
||||
*图 1. 系统监控器发现问题。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
在“资源”面板上显示三个部分：CPU 历史、内存和交换历史，以及网络历史。一眼就能看出你的处理器是否不堪负荷了，或者你的电脑没有内存了，或者你的网络带宽被用光了。
|
||||
|
||||
我将在下面探讨这些问题。现在，当你的电脑速度变慢时，首先检查系统监视器。它可以立即为你提供最常见的性能问题的线索。
|
||||
|
||||
现在让我们来探讨一下如何识别特定领域的瓶颈。
|
||||
|
||||
### 如何识别处理器的瓶颈
|
||||
|
||||
要发现瓶颈，你必须首先知道你有什么硬件。开源为这个目的提供了几个工具。我喜欢 [HardInfo][5]，因为它的屏幕显示很容易阅读，而且广泛流行。
|
||||
|
||||
启动 HardInfo。它的“计算机->摘要”面板可以识别你的 CPU 并告诉你它的核心、线程和速度。它还能识别你的主板和其他计算机部件。
|
||||
|
||||
![HardInfo Summary Panel][6]
|
||||
|
||||
*图 2. HardInfo 显示了硬件细节。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
HardInfo 显示，这台计算机有一个物理 CPU 芯片。该芯片包含两个处理器（或称核心）。每个核心支持两个线程（或称逻辑处理器）。这就是总共四个逻辑处理器 —— 正是图 1 中系统监控器的 CPU 历史部分所显示的。
|
||||
|
||||
当处理器不能在其时间内对请求做出反应时，就会出现 _处理器瓶颈_，说明它们已经很忙了。
|
||||
|
||||
当系统监控器显示逻辑处理器的利用率持续在 80% 或 90% 以上时，你就可以确定这一点。这里有一个例子，四个逻辑处理器中有三个被淹没在 100% 的利用率中。这是一个瓶颈，因为它没有留下多少 CPU 用于其他工作。
|
||||
|
||||
![系统监视器的处理器瓶颈][7]
|
||||
|
||||
*图 3. 一个处理器的瓶颈。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
#### 哪个程序导致了这个问题？
|
||||
|
||||
你需要找出是哪个程序在消耗所有的 CPU。点击系统监视器的“进程”标签。然后点击“CPU 百分比”标头，根据它们消耗的 CPU 的多少对进程进行排序。你将看到哪些应用程序正在扼杀你的系统。
|
||||
|
||||
![系统监控进程面板][8]
|
||||
|
||||
*图 4. 识别违规的进程。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
前三个进程各消耗了 _总 CPU 资源的 24%_。由于有四个逻辑处理器，这意味着每个进程消耗了一整个处理器。这就像图 3 所示。
|
||||
|
||||
在“进程”面板上，一个名为“analytical_AI”的程序被确定为罪魁祸首。你可以在面板上右键单击它，以查看其资源消耗的更多细节，包括内存使用、它所打开的文件、其输入/输出细节，等等。
|
||||
|
||||
如果你的登录有管理员权限，你可以管理这个进程。你可以改变它的优先级，并停止、继续、结束或杀死它。因此，你可以在这里立即解决你的瓶颈问题。
|
||||
|
||||
![系统监视器管理一个进程][9]
|
||||
|
||||
*图 5. 右键点击一个进程来管理它。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
如何解决处理瓶颈问题？除了实时管理违规的进程外，你也可以防止瓶颈的发生。例如，你可以用另一个应用程序来代替违规进程，绕过它，改变你使用该应用程序的行为，将该应用程序安排在非工作时间，解决潜在的内存问题，对该应用程序或你的系统软件进行性能调整，或升级你的硬件。这里涉及的内容太多，所以我将在下一篇文章中探讨这些方式。
|
||||
|
||||
#### 常见的处理器瓶颈
|
||||
|
||||
在用系统监控器监控你的 CPU 时，你会遇到几种常见的瓶颈问题。
|
||||
|
||||
有时一个逻辑处理器出现瓶颈，而其他所有的处理器都处于低利用率。这意味着你有一个应用程序，它的编码不够智能，无法利用一个以上的逻辑处理器，而且它已经把正在使用的那个处理器用完了。这个应用程序完成的时间将比使用更多的处理器要长。但另一方面，至少它能让你的其他处理器腾出手来做其他工作，而不会接管你的电脑。
|
||||
|
||||
你也可能看到一个逻辑处理器永远停留在 100% 的利用率。要么它非常忙，要么是一个进程被挂起了。判断它是否被挂起的方法是，是看该进程是否从不进行任何磁盘活动（正如系统监视器“进程”面板所显示的那样）。
|
||||
|
||||
最后，你可能会注意到，当你所有的处理器都陷入瓶颈时，你的内存也被完全利用了。内存不足的情况有时会导致处理器瓶颈。在这种情况下，你要解决的是根本的内存问题，而不是体现出症状的 CPU 问题。
|
||||
|
||||
### 如何识别内存瓶颈
|
||||
|
||||
鉴于现代 PC 中有大量的内存，内存瓶颈比以前要少得多。然而，如果你运行内存密集型程序，特别是当你的计算机不包含很多随机存取内存（RAM）时，你仍然可能遇到它们。
|
||||
|
||||
Linux [使用内存][10] 既用于程序，也用于缓存磁盘数据。后者加快了磁盘数据的访问速度。Linux 可以在它需要的任何时候回收这些内存供程序使用。
|
||||
|
||||
系统监视器的“资源”面板显示了你的总内存和它被使用的程度。在“进程”面板上，你可以看到单个进程的内存使用情况。
|
||||
|
||||
下面是系统监控器“资源”面板中跟踪总内存使用的部分。
|
||||
|
||||
![系统监控器的内存瓶颈][11]
|
||||
|
||||
*图 6. 一个内存瓶颈。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
在内存的右边，你会注意到 [交换空间][12]。这是 Linux 在内存不足时使用的磁盘空间。它将内存写入磁盘以继续操作，有效地将交换空间作为你的内存的一个较慢的扩展。
|
||||
|
||||
你要注意的两个内存性能问题是：
|
||||
|
||||
1. 内存被大量使用，而且你看到交换空间的活动频繁或不断增加。
|
||||
2. 内存和交换空间都被大量使用。
|
||||
|
||||
情况一意味着更慢的性能，因为交换空间总是比内存更慢。你是否认为这是一个性能问题，取决于许多因素（例如，你的交换空间有多活跃、它的速度、你的预期，等等）。我的看法是，对于现代个人电脑来说，交换空间的任何超过象征性使用都是不可接受的。
|
||||
|
||||
情况二是指内存和交换空间都被大量使用。这是一个 _内存瓶颈_。计算机变得反应迟钝。它甚至可能陷入一种“咆哮”的状态，在这种状态下，除了内存管理之外，它几乎不能完成其他任务。
|
||||
|
||||
上面的图 6 显示了一台只有 2GB 内存的旧电脑。当内存使用量超过 80% 时，系统开始向交换空间写入。响应速度下降了。这张截图显示了内存使用量超过 90%，而且这台电脑已经不可用。
|
||||
|
||||
解决内存问题的最终答案是要么少用内存，要么多买内存。我将在后续文章中讨论解决方案。
|
||||
|
||||
### 如何识别存储瓶颈
|
||||
|
||||
如今的存储有固态和机械硬盘等多个品种。设备接口包括 PCIe、SATA、雷电和 USB。无论你有哪种类型的存储，你都要使用相同的程序来识别磁盘瓶颈。
|
||||
|
||||
从系统监视器开始。它的“进程”面板显示各个进程的输入/输出率。因此，你可以快速识别哪些进程做了最多的磁盘 I/O。
|
||||
|
||||
但该工具并不显示每个磁盘的总数据传输率。你需要查看特定磁盘上的总负载，以确定该磁盘是否是一个存储瓶颈。
|
||||
|
||||
要做到这一点，使用 [atop][13] 命令。它在大多数 Linux 软件库中都有。
|
||||
|
||||
只要在命令行提示符下输入`atop` 即可。下面的输出显示，设备 `sdb` 是 `busy 101%`。很明显，它已经达到了性能极限，限制了你的系统完成工作的速度。
|
||||
|
||||
![atop 磁盘瓶颈][14]
|
||||
|
||||
*图 7. atop 命令识别了一个磁盘瓶颈。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
注意到其中一个 CPU 有 85% 的时间在等待磁盘完成它的工作（`cpu001 w 85%`）。这是典型的存储设备成为瓶颈的情况。事实上，许多人首先看 CPU 的 I/O 等待时间来发现存储瓶颈。
|
||||
|
||||
因此，要想轻松识别存储瓶颈，请使用 `atop` 命令。然后使用系统监视器上的“进程”面板来识别导致瓶颈的各个进程。
|
||||
|
||||
### 如何识别 USB 端口的瓶颈
|
||||
|
||||
有些人整天都在使用他们的 USB 端口。然而，他们从不检查这些端口是否被最佳地使用。无论你是插入外部磁盘、U 盘，还是其他东西，你都要确认你是否从 USB 连接的设备中获得了最大性能。
|
||||
|
||||
这个图表显示了原因。潜在的 USB 数据传输率差异_很大_。
|
||||
|
||||
![USB 标准][15]
|
||||
|
||||
*图 8. (Howard Fosdick，根据 [Tripplite][16] 和 [Wikipedia][17] 提供的数字，[CC BY-SA 4.0][4])的 USB 速度变化很大。*
|
||||
|
||||
HardInfo 的“USB 设备”标签显示了你的计算机支持的 USB 标准。大多数计算机提供不止一种速度。你怎么能知道一个特定端口的速度呢？供应商对它们进行颜色编码，如图表中所示。或者你可以在你的计算机的文档中查找。
|
||||
|
||||
要看到你得到的实际速度，可以通过使用开源的 [GNOME 磁盘][18] 程序进行测试。只要启动 GNOME 磁盘，选择它的“磁盘基准”功能，然后运行一个基准测试。这将告诉你在一个端口插入特定设备时，你将得到的最大实际速度。
|
||||
|
||||
你可能会得到不同的端口传输速度，这取决于你将哪个设备插入它。数据速率取决于端口和设备的特定组合。
|
||||
|
||||
例如，一个可以以 3.1 速度运行的设备如果使用 2.0 端口就会以 2.0 的速度运行。（而且它不会告诉你它是以较慢的速度运行的！）相反，如果你把一个 USB 2.0 设备插入3.1端口，它能工作，但速度是 2.0 的速度。所以要获得快速的 USB，你必须确保端口和设备都支持它。GNOME 磁盘为你提供了验证这一点的方法。
|
||||
|
||||
要确定 USB 的处理瓶颈，使用你对固态和硬盘所做的同样程序。运行 `atop` 命令来发现 USB 存储瓶颈。然后，使用系统监视器来获取违规进程的详细信息。
|
||||
|
||||
### 如何识别互联网带宽瓶颈
|
||||
|
||||
系统监控器的“资源”面板会实时告诉你你所经历的互联网连接速度（见图 1）。
|
||||
|
||||
有 [很好的 Python 工具][19] 可以测试你的最大网速，但你也可以在 [Speedtest][20]、[Fast.com][21] 和 [Speakeasy][22] 等网站进行测试。为了获得最佳结果，关闭所有东西，只运行 _速度测试_；关闭你的 VPN；在一天中的不同时间运行测试；并比较几个测试网站的结果。
|
||||
|
||||
然后将你的结果与你的供应商声称你得到的下载和上传速度进行比较。这样，你就可以确认你得到的是你所付款套餐的速度。
|
||||
|
||||
如果你有一个单独的路由器，在有和没有它的情况下进行测试。这可以告诉你，你的路由器是否是一个瓶颈。如果你使用 WiFi，在有 WiFi 和没有 WiFi 的情况下进行测试（通过将你的笔记本电脑直接与调制解调器连接）。我经常看到人们抱怨他们的互联网供应商，而实际上他们有的只是一个 WiFi 瓶颈，他们可以自己解决。
|
||||
|
||||
如果某些程序正在消耗你的整个互联网连接，你想知道是哪一个。通过使用 `nethogs` 命令找到它。它在大多数软件库中都有。
|
||||
|
||||
有一天，我的系统监视器突然显示我的互联网访问量激增。我只是在命令行中输入了 `nethogs`，它立即确定带宽消耗者是 Clamav 防病毒更新。
|
||||
|
||||
![Nethogs][23]
|
||||
|
||||
*图 9. Nethogs 识别带宽用户。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
### 如何识别图形处理瓶颈
|
||||
|
||||
如果你把显示器插在台式电脑后面的主板上，你就在使用 _板载显卡_。如果你把它插在后面的卡上，你就有一个专门的图形子系统。大多数人称它为 _视频卡_ 或 _显卡_。对于台式电脑来说，附加显卡通常比主板上的显卡更强大、更昂贵。笔记本电脑总是使用板载显卡。
|
||||
|
||||
HardInfo 的“PCI 设备”面板告诉你关于你的图形处理单元（GPU）。它还显示你的专用视频内存的数量（寻找标有“可预取”的内存）。
|
||||
|
||||
![视频芯片组信息][24]
|
||||
|
||||
*图 10. HardInfo提供图形处理信息。(Howard Fosdick, [CC BY-SA 4.0][4])*
|
||||
|
||||
CPU 和 GPU [非常密切地][25]一起工作。简而言之，CPU 为 GPU 准备渲染的帧，然后 GPU 渲染这些帧。
|
||||
|
||||
当你的 CPU 在等待 100% 繁忙的 GPU 时，就会出现 _GPU 瓶颈_。
|
||||
|
||||
为了确定这一点，你需要监控 CPU 和 GPU 的利用率。像 [Conky][26] 和 [Glances][27] 这样的开源监控器，如果它们的扩展插件支持你的图形芯片组，就可以做到这一点。
|
||||
|
||||
看一下 Conky 的这个例子。你可以看到，这个系统有很多可用的 CPU。GPU 只有 25% 的使用率。想象一下，如果这个 GPU 的数量接近 100%。那么你就会知道 CPU 在等待 GPU，你就会有一个 GPU 的瓶颈。
|
||||
|
||||
![Conky CPU 和 GPU 监控][28]
|
||||
|
||||
*图 11. Conky 显示 CPU 和 GPU 的利用率。 (图片来源：[AskUbuntu论坛][29])*
|
||||
|
||||
在某些系统上，你需要一个供应商专属的工具来监控你的 GPU。它们可以从 GitHub 上下载，并在 [GPU 监控和诊断命令行工具][30] 这篇文章中有所描述。
|
||||
|
||||
### 总结
|
||||
|
||||
计算机由一系列集成的硬件资源组成。如果它们中的任何一个在工作量上远远落后于其他资源，就会产生性能瓶颈。这可能会拖累你的整个系统。你需要能够识别和纠正瓶颈，以实现最佳性能。
|
||||
|
||||
不久前，识别瓶颈需要深厚的专业知识。今天的开源 GUI 性能监控器使它变得相当简单。
|
||||
|
||||
在我的下一篇文章中，我将讨论改善你的 Linux 电脑性能的具体方法。同时，请在评论中分享你自己的经验。
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
|
||||
via: https://opensource.com/article/21/3/linux-performance-bottlenecks
|
||||
|
||||
作者：[Howard Fosdick][a]
|
||||
选题：[lujun9972][b]
|
||||
译者：[wxy](https://github.com/wxy)
|
||||
校对：[校对者ID](https://github.com/校对者ID)
|
||||
|
||||
本文由 [LCTT](https://github.com/LCTT/TranslateProject) 原创编译，[Linux中国](https://linux.cn/) 荣誉推出
|
||||
|
||||
[a]: https://opensource.com/users/howtech
|
||||
[b]: https://github.com/lujun9972
|
||||
[1]: https://opensource.com/sites/default/files/styles/image-full-size/public/lead-images/BUSINESS_lightning.png?itok=wRzjWIlm (Lightning in a bottle)
|
||||
[2]: https://wiki.gnome.org/Apps/SystemMonitor
|
||||
[3]: https://opensource.com/sites/default/files/uploads/1_system_monitor_resources_panel.jpg (System Monitor - Resources Panel )
|
||||
[4]: https://creativecommons.org/licenses/by-sa/4.0/
|
||||
[5]: https://itsfoss.com/hardinfo/
|
||||
[6]: https://opensource.com/sites/default/files/uploads/2_hardinfo_summary_panel.jpg (HardInfo Summary Panel)
|
||||
[7]: https://opensource.com/sites/default/files/uploads/3_system_monitor_100_processor_utilization.jpg (System Monitor processor bottleneck)
|
||||
[8]: https://opensource.com/sites/default/files/uploads/4_system_monitor_processes_panel.jpg (System Monitor Processes panel)
|
||||
[9]: https://opensource.com/sites/default/files/uploads/5_system_monitor_manage_a_process.jpg (System Monitor managing a process)
|
||||
[10]: https://www.networkworld.com/article/3394603/when-to-be-concerned-about-memory-levels-on-linux.html
|
||||
[11]: https://opensource.com/sites/default/files/uploads/6_system_monitor_out_of_memory.jpg (System Monitor memory bottleneck)
|
||||
[12]: https://opensource.com/article/18/9/swap-space-linux-systems
|
||||
[13]: https://opensource.com/life/16/2/open-source-tools-system-monitoring
|
||||
[14]: https://opensource.com/sites/default/files/uploads/7_atop_storage_bottleneck.jpg (atop disk bottleneck)
|
||||
[15]: https://opensource.com/sites/default/files/uploads/8_usb_standards_speeds.jpg (USB standards)
|
||||
[16]: https://www.samsung.com/us/computing/memory-storage/solid-state-drives/
|
||||
[17]: https://en.wikipedia.org/wiki/USB
|
||||
[18]: https://wiki.gnome.org/Apps/Disks
|
||||
[19]: https://opensource.com/article/20/1/internet-speed-tests
|
||||
[20]: https://www.speedtest.net/
|
||||
[21]: https://fast.com/
|
||||
[22]: https://www.speakeasy.net/speedtest/
|
||||
[23]: https://opensource.com/sites/default/files/uploads/9_nethogs_bandwidth_consumers.jpg (Nethogs)
|
||||
[24]: https://opensource.com/sites/default/files/uploads/10_hardinfo_video_card_information.jpg (Video Chipset Information)
|
||||
[25]: https://www.wepc.com/tips/cpu-gpu-bottleneck/
|
||||
[26]: https://itsfoss.com/conky-gui-ubuntu-1304/
|
||||
[27]: https://opensource.com/article/19/11/monitoring-linux-glances
|
||||
[28]: https://opensource.com/sites/default/files/uploads/11_conky_cpu_and_gup_monitoring.jpg (Conky CPU and GPU monitoring)
|
||||
[29]: https://askubuntu.com/questions/387594/how-to-measure-gpu-usage
|
||||
[30]: https://www.cyberciti.biz/open-source/command-line-hacks/linux-gpu-monitoring-and-diagnostic-commands/
|
||||
Loading…
Reference in New Issue
Block a user