Weekly Bullet #24 – Summary for the week

Here are a bunch of Technical / Non-Technical topics that I came across recently and found them very resourceful.

Technical :

  • An overview of iftop – a great network traffic visual tool. – here , also man page here
  • Rust is becoming one of the most loved languages. Here is an Illustrated Note about WTF is Rust – link
  • How They SRE” – best practices, tools, techniques, and culture of SRE adopted by the leading technology or tech-savvy organizations.- link
  • A single stop to find all upcoming Tech conferences in 2021 – link
  • A wiki on Unix Toolbox with all the commands and tasks useful for daily dive in to linux world. – link
  • “Python Tricks I cannot live without” – link
  • I am sure most of you follow HackerNews. Here is a great tool built using FlameGraphs to navigate through big threads on HN. – Link1 , Link2

Non-Technical :

  • A cool site where you can select the part of the body and find the relevant stretches and exercises here
  • An extract from something I am reading:

“Almost universally, the kind of performance we give on social media is positive. It’s more “Let me tell you how well things are going. Look how great I am.” It’s rarely the truth: “I’m scared. I’m struggling. I don’t know.”

Ryan Holiday, Ego Is the Enemy

Have a good week ahead.

[Performance] : Using iperf3 tool for Network throughput test

In this world of Microservices and the distributed systems, a single request (generally) hops through multiple servers before being served. More often than not, these hops are also across the Network cards making the Network Performance the source of slowness in the application.
These parameters makes the need to measure Network performance between servers/systems more critical for benchmarking or debugging.

Iperf3 is one of the open source tools which can be used for network throughput measurement. Below are some of its features.

  • Iperf3 can be used for testing maximum TCP and UDP throughput  between two servers.
  • Iperf3 tests can also be run in a controlled to way to not test the maximum limits but ingest and constant lower network traffic for testing.
  • Iperf3 has options for parallel mode(-P) where multiple clients can be used, setting CPU affinity(-A), pausing certain intervals between two requests(-i), setting the length of buffer to read or write(-l), setting target bandwidth (-b) etc.
  • More important than anything is the fact that iperf3 runs as an independent tool outside your application code. The results from this tool removes any ambiguities/doubts on the application code which might be causing the network problems.

Installation of iperf3 tool:

sudo apt-get install iperf3	

iperf3 tool has be installed on both servers between which you want to measure the network performance. One of the machines is treated as client and other as server.

Command to run on the server:

Below command when run on one of the two servers under test, signifies that the machine is acting as a server for the iperf test.

iperf3 -s -f K
  • -s — runs in server mode
  • -f K — signifies the format as KBytes.
    Note : If you do not want to use the default port (which is 5201) for the test, then specify the port with the option -p in the above command and use the same on client as well.

Command to run on the client:

Below command when run on the other server under test, pushes network bandwidth to server and reports the network capacity based on options used.

iperf3 -c 192.XX.XX.XX -f K
==== output ====

Connecting to host 192.XX.XX.XX, port 5201
[  4] local 192.XX.XX.XX port 50880 connected to 192.XX.XX.XX port 5201
[ ID] Interval           Transfer     Bandwidth       Retr  Cwnd
[  4]   0.00-1.00   sec   678 MBytes  693729 KBytes/sec   74   1.06 MBytes
[  4]   1.00-2.00   sec   750 MBytes  767998 KBytes/sec    0   1.48 MBytes       
[  4]   2.00-3.00   sec   606 MBytes  620723 KBytes/sec  143   1.22 MBytes       
[  4]   3.00-4.00   sec   661 MBytes  677201 KBytes/sec    0   1.57 MBytes       
[  4]   4.00-5.00   sec   620 MBytes  634523 KBytes/sec    0   1.83 MBytes       
[  4]   5.00-6.00   sec   609 MBytes  623718 KBytes/sec  1095   1.44 MBytes       
[  4]   6.00-7.00   sec   730 MBytes  747525 KBytes/sec    0   1.76 MBytes       
[  4]   7.00-8.00   sec   716 MBytes  733224 KBytes/sec    0   2.04 MBytes       
[  4]   8.00-9.00   sec   772 MBytes  791192 KBytes/sec    0   2.29 MBytes       
[  4]   9.00-10.00  sec   944 MBytes  966472 KBytes/sec  212   1.63 MBytes       
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec  6.92 GBytes  725627 KBytes/sec  1524 sender
[  4]   0.00-10.00  sec  6.92 GBytes  725350 KBytes/sec       receiver
  • -c — run in client mode
  • The ip mentioned is the ip of the server.
  • From the output (last two lines), it can be seen that the total bandwidth available between the two servers is 708MBytes/sec.

There are also various other options available for iperf3 tool. Like the below command specifies the test to be run for 60secs, which is by default 10secs (-t 60), specifies a target bandwidth of 10MB (-b 10M), number of parallel client streams set to 10 (-P 10).

iperf3 -c 192.XX.XX.XX -t 60 -b 10M -P 10 --get-server-output
  • –get-server-output : this get the command line output from server and prints it on client terminal
  • If you want to use udp instead of tcp, same can be achieved using the option –u
  • More details available here on man page – link

Below are some of the cases where I have used iperf3 for debugging purpose:

  • Throughput of the application doesn’t scale but there is no obvious resource contention in cpu, memory or disk. On running “sar -n DEV 2 20” I could see network usage doesn’t peak above 30MB/sec. On using Iperf3 for benchmarking we could see 30Mb/sec was the max network capacity between the servers.
  • When we wanted to find the impact of vpn on the network throughput we used iperf tool for comparative analysis.

Hope this gave you a sneak peak into iperf3 tool’s capability and usages.
Happy tuning!

[Performance] : Java Thread Dumps – Part2

In the previous article about Java Thread Dumps (link here) we looked in to a few basics on Thread dumps(When to take?, How to take?, Sneak peaks? etc.)

In this write up, I wanted to mention a few tools which can ease the process of collecting and analyzing thread dumps.

Collecting multiple thread dumps:

I prefer command-line over any APM tools for taking thread dumps. The best way for analyzing threads is to collect a few thread dumps (5 to 10) and look through the transition in the state of threads.

As mentioned in the previous article(link), one of the ways is using jstack, which is built in to jdk. Below command will collect 10 thread dumps with a time interval of 10sec between each. All dumps are written to a single file ThreadDump.jstack.

a=1; while [ $a -le 10 ]; do jstack -l <pid> >> ThreadDump.jstack && sleep 10; a=`expr $a + 1`; done

You can further split the 10 thread dumps to individual files using the csplit command. The below command basically looks for the line “Full thread dump OpenJDK” which is printed at the started of each thread dump and splits the dump in to individual ones.

csplit -z ThreadDump.jstack /Full\ thread\ dump\ OpenJDK\ 64-Bit\ Server/ '{*}' 
ThreadDumps split into individual files

Tools for visualizing and analyzing ThreadDumps:

To start with, there are many tools like VisualVM, Jprofiler, Yourkit and many more online aggregators for visualizing and analyzing ThreadDumps. Each one have their own pros & cons.
Here is a small list of available tools – link

I generally use the “Java Thread Dump Analyser” (JTDA) tool from here.

This tool is available as a jar, to which you can feed the Thread dump file via command line and visualize Threads grouped in there States. You would appreciate this grouping if you have analyzed Thread Dumps in vim which had 100’s and 100’s of lines.

java -jar jtda.jar <ThreadDumpFilePath>

Here are some of the features which I like about JTDA tool:

  • light weight and doesn’t need any configuration.
  • you can upfront get the count of threads in all states
  • you can only look in to threads that are in a certain state (Runnable / Waiting / Blocked etc)
  • threads with same stack are grouped together for ease of reading
  • “show thread name” if checked, give the name of thread pool for better context.

On the foot note, when looking in to thread dumps, it is very important to know the code paths and the request flows. This helps in root causing the issue and better reading / understanding of thread dumps.

Happy tuning.

Weekly Bullet #23 – Summary for the week

Here are a bunch of Technical / Non-Technical topics that I came across recently and found them very resourceful.

Technical :

  • BPF(Berkeley Packet Filter) has come a long way from just being a packet capture tool to advance Performance analysis tool (EBPF – Extended Berkeley Packet Filter). Here (link) is an introduction to EBPF. Also here (link) is a talk on how BPF is used at Netflix.
  • “Minimal safe Bash script template” – link . Because there is no such thing as “knowing enough of bash!”
  • Kelsey Hightower is an inspiration. A writeup on how he made it from McDonald’s to Google (link). [HIGHLY RECOMMENDED] –> : A talk he gave about his journey a few years back here (link)
  • [That time of the year!] : “Best talks of 2020” — link
  • [Late news!] If you didn’t hear it already, Github has Dark mode now. – link

Non-Technical :

  • [Another one] “Ask HN: What book changed your life in 2020?” – some great recommendations here – link . Personally for me, “Sapiens” widened my horizon about evolution of Human Beings.
  • “100 Tips for better life.” – link – I don’t agree with all of them, but most of these are thought provoking.
  • An extract from the book that I am reading.

At the core of all anger is a need that is not being fulfilled.

Marshall B. Rosenberg, Nonviolent Communication: A Language of Life

Happy learning and an advance Happy new year 2021!

Weekly Bullet #22 – Summary for the week

Here are a bunch of Technical / Non-Technical topics that I came across recently and found them very resourceful.

Technical :

  • [Talk-Velocity 2017] : Performance Analysis Superpowers with Linux eBPF (44mins)- link
  • Popular Java Podcasts to follow in 2020 – link
  • Since we are taking about Podcasts, I have also heard good things about Barcode and ACM-ByteCast is amazing!
  • Illustration: Much that we have gotten wrong about SRE – link
  • A list of popular java libraries. – link
  • The second edition of “System Performance: Enterprise and Cloud” – by Brendan Gregg releasing on 2nd December. – link . This is “the best” reference guide for Performance Engineering.

Non-Technical :

  • “Library of Scroll” – Here is a site with one great article every Monday. Since it is just one, generally I find them very good. – link
  • Great site with short explanations of over 24 cognitive biases. Co-authored by Gabriel Weinberg who is the CEO of DuckDuckGo. – link
  • Not sure why I liked this, but this “57 Years Apart – A Boy And a Man Talk About Life” short video was quite gripping. – link
  • Soft skills for Software Engineers. Short thread. – link
  • Hand picked remote jobs from “Hacker News Who is hiring” November – link
  • Extract from a book:

“Respect an old tradition path as it is well tested, but also be open to the new modern way of things as they open up your mind.”

The Daily Stoic

Have a great week ahead.

Thinking in-terms of Performance

A few short thoughts / ideas wrt of Performance centric product.

  • In this world of infinite scaling of computes, pay close attention to common chock points. Like DB, storage(s) etc, which are shared by all the computes.
  • Majority of the reads and writes have to happen in Bulk operations and NOT as single read/writes. Specially when there are 100’s-1000’s of reads/writes/deletes on storage(s).
  • Threads. Pay close attention to which part of the entire flow is multi-threaded. Sometimes, only a small part of the flow is multi-threaded, but entire application is called multi-threaded, which is wrong.
  • Don’t just look at throughput for certifying if an application is performing good / bad.
    Scenario 1: Throughput:100docs/sec, Object size:100Kb
    Scenario 2: Throughput:50docs/sec, Object size:200Kb
    In both the cases about the amount of data transferred on network is 10,000Kb/sec (9.7Mb/sec). So pay attention to network stats and size of objects as well.
  • Databases. Make sure important collections have indexes. (There is so much more to DB tuning. This could be starting point)
  • Introduce local cache on computes wherever possible to avoid network calls to DB’s. Be 100% sure on your logic on when to invalidate the cache. If we don’t invalidate the cache and use the stale cached data, that could lead to incorrect transactions.
  • Failures are important. Pay close attention to reducing the failures. If something has to fail, then fail early on its compute cycle, so that CPU cycles are not wasted on failures.
  • When you do retries due to failures in your system, make sure to have logarithmic back offs. Meaning, if a message failed, 1st retry after 1min, 2nd retry after 2mins, 3rd after 4mins etc. This give system time to recover. Also retry only for finite number of times.
  • Pay very close attention on logic for when to mark a failure as Temporary vs Permanent. If a failure which should be permanent is wrongly marked as temporary, then the request keeps coming back to the compute via retries and wastes the CPU cycles.
  • Timeouts and Max connection pool sizes! Make a note of all the timeout that are set across talks between different fabrics / api calls in your application. Also, see the configuration for connections pools used in the application. Both might result in spike of failures if crossed.

Happy to converse about any in the comments.

Weekly Bullet #21 – Summary for the week

Here are a bunch of Technical / Non-Technical topics that I came across recently and found them very resourceful.

Technical :

  • How to become a consultant ? Some good references and advices here. – link
  • Ever wondered How the prices vary on Amazon? Here is a classic example of algorithmically priced products on Amazon – link
  • If you are using O’Reilly (which I believe is the best technical content platform), you should check out O’Reilly Answers. For all your queries, O’Reilly looks through heap of books, video and conferences and gives you answers. – link (Note: You will need subscription)
  • In Firefox version 81, an experimental event delay tracker has been added. Details in the “Performance Tools” section of below article (old article dated August 31st). – link . Also example profile – link
  • An extensive collection of Bash pitfalls – link

Non-Technical :

  • As “Arguments” are a part of corporate jobs, here is a Beginner’s Guide to Arguing Constructively. – link
  • Expiring vs Permanent skills. Many would agree that these skills are more important than absolute technical skills. – link
  • Nokia is going to build a mobile network on Moon – link
  • If you are into living in small spaces and minimalism, you will love this youtube channel, “Never too small”. link . Also, my recent favorite – link
  • [Repost, because why not!] – This never gets old. Richard Feynman’s – “Names Don’t Constitute Knowledge” principle (2mins) – link
  • An extract from a book that I am reading:

Fooling with books so you can sound smart or have an intimidating library is like tending a garden to impress your neighbors.

The Daily Stoic

[Performance] : Java Thread Dumps – Part1

This is first of a two parts article which talks about:

  • What are thread dumps?
  • When to take thread dumps ?
  • How to take thread dumps ?
  • What is inside a thread dumps ?
  • What to look for in a thread dump?

Majority of the systems today are mutlicore and hyper-threaded. Threading at the software level allows us to take advantage of a system’s mutlicores to achieve the desired pace and efficiency of the application operations. Along with pace and efficiency, multi-threading brings its own set of problems w.r.t thread contentions, thread racing, high CPU usage etc. In this write up we will see how to debug these problems by taking thread dumps on java applications.

What are thread dumps ?

A thread dump is a runtime state of a program at a particular point in time. It can also be looked at as a snapshot of all the threads that are alive on the JVM. In a mutli-threaded java application there can be 100’s of threads. Although mutli-threading improves the efficiency of an application, it makes it complex to know what is happening at a given point in the application.
This is where thread dumps come in handy, which gives us a textual representation of all the thread stacks of all the threads in the application.

When to take thread dumps ?

Thread dumps can be helpful in the scenarios like:
– you want to know the state of application.
– you want to know if all the threads from the assigned thread pool are used and are doing work.
– when you are seeing high CPU and want to see which current running threads are causing high cpu.
– when you are seeing unusual slowness in the application.
– with certain jvm parameters, jvm takes auto thread dumps if jvm crashes to help debugging the crash.

How to take thread dumps ?

Two easy ways of taking thread dumps are using jstack or with kill -3 command.

It is best to take a bunch of thread dumps with a slight time gap to better analyse the application. If you use jstack, you could take 10thread dumps with 10seconds sleep time between each, using below shell command. Note: replace <pid> with your process id in below command.

a=1; while [ $a -le 10 ]; do jstack -l <pid> >> bigdump.jstack && sleep 10; a=`expr $a + 1`; done


There is enough content available online on these. So I will link them below.
– Oracle documentation on jstack link
– Redhat knowledge base link
– Tutorial point link
– Multiple other ways link
– If you are using kill -3, note that thread dumps will go to system out of JVM – link

What is inside a thread dump ?

Now that we are done with boilerplate sections, lets see the main part.
When you open a thread dump, you generally see three kinds of threads.
JVM threads(some), Application threads(many) and Heap/GC threads(few). Snippets of these threads are shown below.

Three kinds of threads in ThreadDumps

Elements in a single thread stack:

For debugging an application, Application threads are our area of interest. When inspecting a single Application thread, it has below parts in it.

components of a single application thread

1 – Thread name, obvious and straight forward .
2 – Number id of thread. This will get incremented everytime there is a new thread created for that thread pool.
3 – JVM priority. This number signifies the priority of the thread in JVM. Java exposes api for setting thread priority.
4 – Priority of a thread on the operating system.
5 – Thread id for the thread in the memory
6 – Native thread id assigned to JVM by OS. This is important of correlating JVM threads with OS threads.
7 – This is optional seen only when the thread is waiting on some condition. May not be always.
8 – Thread state. In my opinion, this is overrated and generally over-analyzed. Reflect the current state of the thread.
9 – Thread call stack. Reads bottom-up. Gives the stack of thread.

States of a thread:

This is element [8] in the above highlighted breakdown.
Below are the different states that a thread might be in.

RUNNABLE, WAITING, TIMED_WAITING, BLOCKED, NEW, TERMINATED

  • RUNNABLE – This is the thread which is running on CPU and actively executing code.
  • WAITING – A threads in this state will wait indefinitely unless some other thread wakes it up. Some examples are threads which are waiting on IO or threads waiting for data to be returned from DB.
  • TIMED_WAITING – Similar to waiting but the thread will wake up after a specific period of time.  Storm Spout threads which wake up every few seconds are a good example.
  • BLOCKED – Threads waiting to acquire a monitor lock. In Java, blocks of code can be marked as “synchronized”, so when a thread acquires a monitor lock, other threads will wait in this state, until the monitor is available.
  • NEW & TERMINATED – these states are rarely seen and are of least interest.

What to look for in a thread dump?

Now that we know most part of what can be found in a thread dump, lets see what we should be looking at for analysis.

  • The first important thing is to look at your application threads related info to begin with. For this, pay attention to thread name ( Element [1] in above breakdown). Obviously one would know the thread pools created by the application, and you need to look at the threads whose name belong to this Thread pool names.
  • Look out for the threads in RUNNABLE state amongst the above selected application threads. These are the threads which are actually doing the work and running on the cpu.
    Example: If a thread pool of 200 is set and if only 10 of those threads are in RUNNABLE state, then you might want to look at what rest of the 190 threads are doing.
  • Even in the RUNNABLE threads, it is important to see if the thread is busy executing any NATIVE methods. When you see the calls being made to Native methods in the thread dump, that means the thread is not executing Application code at that point, but is executing JNI (java native interface) code.
    Native methods are pre-compiled piece of codes from JVM which might be serving small specific purposes like reading from socket connections (like in below example), making file system calls etc. If you see a lot of threads in RUNNABLE state, but are busy executing Native methods, then it might not be application code issue but environment issue in itself. More often, some other tools like sTrace are coupled with Thread dumps to root cause these issues, which we shall discuss in next article.
An application thread in Runnable state executing Native method
  • As mentioned in the section, How to take thread dumps, generally it is good to take a few thread dumps (like 10) to understand the state of threads better. These are particularly helpful in looking at code stagnation. Since thread dumps are snap shots of current state of thread, you might want to see if a particular thread that you are interested in (if in Waiting state), is in the same state across all the dumps collected.
  • See if any threads are in BLOCKED state. From the stack trace of the thread, look for the part which says – “waiting to lock <_id here>“. On searching the entire ThreadDump with the same thread_id, you see get to know the thread which is holding the lock and also the problematic object, which in below example is threadstates.Foo.
Example code from the internet
  • The stack trace of threads are important as they help you correlate the problem to the code sections from your application. So it is important to pay close attention to the stack of a thread.

In the next part we will see some practical cases of using Thread dump for analysis, Tools that can be used for analyzing thread dumps.

Happy tuning!

PS: Part 2 of the series is available here

[Performance] : Flame Graphs

In the previous article we explored the basic capabilities of linux Perf_tool.
In this write-up I am trying to extend these capabilities and show how to generate and read Flame Graphs for analyzing the profiles generated with Perf_tool.

How to generate Flame Graphs ?

  • To start with, we will need perf_tools linux profiler to capture the profile first. Follow the steps under “How to setup perf tool?” in the previous article.
  • Now if you collect a profile of the CPU using perf_tool setup in the above step, there is a possibility that you might see a lot of symbol link values in the place of Function names.
  • To avoid this and keep the names of functions for the corresponding Symbol links resolved before hand, clone the below repository on the machine which you want to profile and follow below steps.
* git clone https://github.com/jvm-profiling-tools/perf-map-agent.git
* Navigate to installation directory where there is "CMakeLists.txt" in the above repo
* run " cmake ."
* run "make"
* Navigate to the directory /perf-map-agent/bin in the above repository
* run "./create-java-perf-map.sh <pid>" --> where <pid> is the processId of the process you want to profile
* This will generate a "perf-<pid>.map" file in /tmp -- this will be used for resolving all the symbol links
  • Clone the Flame Graph dependencies from the below repository.
    git clone https://github.com/brendangregg/FlameGraph
  • Alright, almost there! Now lets collect the CPU profile for the process which we want to investigate.
    Note : More details about the below command and the capabilities of perf_tool in the previous article.
perf record -F 99 -ag -p PID-- sleep 60
  • Now lets generate a flame graph out of the profile collected.
#Copy the perf.data file generated from the profiler in the above step
perf script | ./stackcollapse.pl > outperf-folded
./flamegraph.pl outperf-folded > flameGraph.svg
  • And open the flameGraph.svg in one of the browsers.

How to read a Flame Graph ?

  • Now that we have successfully generated the flame graphs, lets see how we can use it for analysis.
  • Flame Graphs are read bottoms up, just like a Thread dump.
  • If you use the parameter "--color=java” while generating the FlameGraph.svg file, you will get different color codes for different type of frames like Green is Java, yellow is C++, orange is kernel, and red is the remainder (native user-level, or kernel modules).
source : Brenden Gregg’s Flame Graph page
  • In the flame graph, the length of each tile (frame) represents the time it took on cpu. That means, larger the length of a tile, larger was its CPU time. So while debugging an issue, you can pick a larger tile and use it as a starting point for debugging.
  • The stack of functions on the base tile represent the child calls from within the base function. Using this we can see if any non-required / repetitive calls are made to any functions.
  • Also, all flame graphs are interactive. You can click on any function tile and see what % of time was spent on it, number of samples collected and what are the child calls with in.
  • On a whole, Flame Graphs are a great tool to generate interactive profiles to see where most of the CPU time is spent in the code.
  • Although setting up prerequisite & generating FlameGraphs has a steep learning curve, I would say it is a great tool to have in a Performance Engineer’s arsenal.
  • It pays off when you want to look at all the system level (low-level kernel, io etc) and application level (through all stacks) cpu calls in a single place. Most of the other profilers fall short at this.

How it helped me solve a Performance Issue ?

  • I had an issue at hand where the system under test (storm) had high cpu usage even when there was no load.
  • On using Yourkit, I did not get anything in the HotSpot other than some threads which were getting on CPU and then going back to sleep state.
  • On looking at FlameGraph however, I could see that these threads which are periodically waking up are Storm Spouts, that are making a lot of system calls via vDSO(virtual dynamic shared object) to get the clocktime of the system.
  • This guided me to check the clocksource set on the system, which was xen and then change it to more optimum tsc.
  • Overall helping us save over 30% of CPU time.

Further reading :

Happy tuning.

[Performance] : Profiling with linux Perf command-line tool

Most of the Performance Engineers use some sort of profiling tools like Yourkit, Jprofiler or some APM tools like Newrelic, Datadog, Appdynmics etc. Although these tools are easy to use out of the box and help with Observability, they don’t give a complete picture of a Performance problem at occasions.
This is where perf Linux profiler comes in handy.

This write up is an attempt to explain :
– What is perf Linux profiler ?
– How to set it up ?
– What are its capabilities ?

So, What is perf linux profiler ?

  • perf Linux profiler also known as “Performance Counter for Linux” (PCL), “Linux Perf Events” (LPE) or “perf_events”, is an incredibly powerful tool for Linux Performance investigations. In this article we will refer to it as just “perf tool“.
  • perf tool is a multi-tool which is capable of both Sampling and Tracing. But unlike the traditional profiling tools, perf tool will trace the system calls as well and not just the application calls. This is greatly helpful in debug is system level issues.
  • Using perf tool we can instrument CPU performance counters, tracepoints and do dynamic tracing.
  • From the profile data generated from perf tool, we can create Flame Graphs, which give an amazing insights in to CPU time and call stacks. A detailed article on Flame Graphs is coming in the following weeks.
sample flame graph

How to set up perf ?

  • perf tool is built in to linux kernel tree.
  • However, if you don’t find it in your linux distribution, you can install it like below on Ubuntu machine :
    sudo apt-get install linux-tools-common
  • perf tool is dependent on the kernel version on your system. So if you get any errors, first check the version of you kernel and then install the relevant linux-tools-common version.
    sudo apt-get install linux-tools-common-$(uname -r)
  • Based on the missing dependencies, after above command, it might ask you install a specific version of linux-cloud-tools, do the same.
  • To check if perf tools is installed successfully, just run the command perf. If it returns the output, you are all set !

What are the capabilities of perf linux profiler ?

Things you can do with perf tool

perf tool has way more capabilities than any of the traditional profiling tools. There is too much of information that we can extract out of a system using perf tool, but it is useful only if we know what to make out of it. Using perf tool we can do things like : profile a PID, get the counter stats on a system, get details on reason for context switches & page faults on a system, collect traces for a pid, do dynamic tracing and much more. Let us look a few of them below.

perf stat :

perf stat is used for obtaining performance counter statistics. This can be used for keeping a running count of an execution, which is aggregated in the end and is presented in a output.
* perf stat --help : To get all the options that can be used with stat
* perf stat -d command : To get CPU counter statistics for any command. “-d” gives detailed results.
* perf stat -p PID sleep 10 : To get stats for a particular PID, for 10secs.
* perf stat -e ext4:ext4_request_inode -a sleep 10 : To check the no. of requests for inodes.ext4 file system provides around one hundred tracepoints for visibility into its requests and internals.
* perf stat -e context-switches -a sleep 5 : To get total number of context switches across all CPUs( “-a”). Also, “-e” option is for event selector. use ‘perf list‘ to list available events.
* Using pref stat you can get a lot more information on cpu cache hits / misses, stalled cycles, cpu-migrations etc.

output of perf stat command

perf top :

perf top is a system profiling utility. This command generates and displays a performance counter profile in real time.
* perf top -F 49 : To sample CPUs at 49 Hertz, and show top addresses and symbols, live. “-F” signifies the frequency for profiling.
* Just like the options mentioned in perf stat above, we can use options for getting results for a particular command, for a PID or for a particular event type.

samples collected with perf top

perf record :

perf record is used to collect the system wide statistics. It is basically static tracing, which can be used on a command, pid or particular event type to get the detailed static trace of the same. This is more like what you see on collecting a trace in Yourkit, but along with all the detailed kernel calls. Also, unlike Yourkit where you can collect only CPU traces, using perf record you can collect traces for all system events like context switches, disk IO, inbound/outbound calls etc.
perf record generates a perf.data file, which is consumed by perf reports commands for showing out the details.
* perf record -e sched:sched_process_exec -a sleep 20 : To trace all the new processes for the next 20secs.
* perf record -e context-switches -ag -- sleep 10 : To sample context switching for 10 seconds. “-a” means on all CPUs. “-g” means enabling call graph recording, which will generate detailed stacks in report.
* perf record -F 99 -p PID sleep 10 : To record all the instructions that go on CPU at the frequency of 99Hertz for a specific PID for 10secs.

All the above recordings will generate a perf.data file which is used as input for perf report command for analyzing the recorded samples.

output after running a perf record command

perf report :

The profile data collected by perf record command are saved in perf.data binary file as seen above. perf report reads this binary file and create a report out of it.
* perf report -n : To show the profile data, “-n” shows the details of number of samples collected for each function.

perf report -n

* perf report -n --stdio : To generate the report with stdio interface.

perf report -n –stdio

Side note : Make sure to “overuse” the manual pages in perf tool. It is super helpful in listing all the options that you can use with perf commands. Use --help along with any perf command to get the manual page.

Although this write is an attempt to introduce someone to perf linux profiler, it has far too many capabilities than that can be listed in a single write-up. The followup articles will attempt to get deeper in to perf tool features & generation of Flame Graphs out of perf reports.

Happy tuning!

Important links :